REPUBLIC OF TURKEY FIRAT UNIVERSITY
THE GRADUATE SCHOOL OF NATURAL AND APPLIED SCIENCE
FRIEND RECOMMENDATION SYSTEM IN ONLINE SOCIAL NETWORKS
Mohammed adam Faris MOHAMMED (142129106)
Master Thesis
Department: Computer Engineering Supervisor: Prof. Dr. Mehemt KAYA
JANUARY- 2017
REPUBLIC OF TURKEY FIRAT UNIVERSITY
THE GRADUATE SCHOOL OF NATURAL AND APPLIED SCIENCE
FRIEND RECOMMENDATION SYSTEM IN ONLINE SOCIAL NETWORKS
Department of Computer Engineering
Master Thesis
Mohammed adam Faris MOHAMMED (142129106)
Advisor of the Thesis Prof. Dr. Mehmet KAYA
REPUBLIC OF TURKEY FIRAT UNIVERSITY
THE GRADUATE SCHOOL OF NATURAL AND APPLIED SCIENCE
ACKNOWLEDGMENT
Thanks God before everything and after everything for giving me the knowledge and ability to complete this research work in this final form.
I am very thankful and indebted to my respected teacher Prof. Dr. Mehmet KAYA for his valuable and thorough supervision throughout all the phases of my thesis, without which it wouldn’t be possible to be completed. Not just during my thesis preparation but from the time when I came to Turkey he is one of the persons who helped me very much.
My special thanks goes to Asst. Prof. Dr. Taner TUNCER and Ahmet Anil Mugen and for taking part as an advisory committee in my thesis presentation and their inestimable feedbacks which enhanced and improved my research.
I would like to express my gratitude and special thanks to Turkey Government and Presidency for Turks aboard and related communities (YTB (Yurtdışı Türkler ve Akraba Topluluklar Başkanlığı)) for providing the master degree scholarship to me, by which I found the ability to become familiar to Turkish people, Turkish culture. Their unlimited helps, supports and encouragements are greatly appreciated.
Acknowledging my beloved family for their supports and encouragements in the hard times, I am forever indebted to my family especially my mother and my father and my wife for all their helps both materially and morally.
My special thanks goes to my dear friends and all faculty members for their helps.
Mohammed adam Faris MOHAMMED ELAZIG 2017
CONTENTS ACKNOWLEDGMENT _________________________________________________________ I CONTENTS ___________________________________________________________________ II LIST OF FIGURES ____________________________________________________________ IV LIST OF TABLES _____________________________________________________________ VI ABSTRACT _________________________________________________________________ VII ÖZET _____________________________________________________________________ VIII 1. INTRODUCTION ____________________________________________________________ 1 1.1. Related Work _______________________________________________________________ 2 1.2. Organization of The Thesis _____________________________________________________ 6
2. SOCIAL NETWORKS ________________________________________________________ 7
2.1. Social Networks _____________________________________________________________ 7 2.2. Online Social Networks ______________________________________________________ 10 2.3. Functions of Online Social Network Systems _____________________________________ 12 2.4. Features of Social Networks ___________________________________________________ 13 2.5. Top 10 Social Networks ______________________________________________________ 14 2.6. Social Network Analysis ______________________________________________________ 15 2.6.1. Basic Metrics in Social Network Analysis ______________________________________ 16 2.7. Related Work ______________________________________________________________ 20
3. RECOMMENDATION METHODS ____________________________________________ 25
3.1. Definition of Recommender Systems ____________________________________________ 26 3.2. Goals of Recommender Systems _______________________________________________ 27 3.3. Categories of Recommender Systems ____________________________________________ 28 3.3.1. Collaborative Filtering: ____________________________________________________ 32 3.3.2. Content-Based Filtering ___________________________________________________ 33 3.3.3. Hybrid Method __________________________________________________________ 34 3.4. Related Work ______________________________________________________________ 34
4. LINK PREDICTION FOR RECOMMENDATION _______________________________ 35
4.1. Link Prediction _____________________________________________________________ 35 4.2. Traditional Proximity Measures ________________________________________________ 36 4.2.1 Number of Common Neighbors (CN) __________________________________________ 36 4.2.2 Jacquard’s coefficient ______________________________________________________ 36 4.2.3 Preferential Attachment (PA) ________________________________________________ 37
5. EXPERIMENTAL RESULT __________________________________________________ 38
6. CONCLUSIONS AND FUTURE WORKS _______________________________________ 44
REFERENCE _________________________________________________________________ 45 CURRICULUM VITAE ________________________________________________________ 50
LIST OF FIGURES
Figure 1.1. Social Network _________________________________________________ 1 Figure 2.1. The division of social networks based on type of relationship _____________ 9 Figure 2.2. The division of social networks based on type of communication channel ___ 10 Figure 2.3. The The taxonomy of CSSN. ______________________________________ 11 Figure 2.4. Ranked number of active users in million (2016). ______________________ 15 Figure 2.5. An example network for degree centrality calculation __________________ 17 Figure 2.6. An example network for closeness centrality calculation ________________ 18 Figure 3.1. Levels of the personalization in the recommender systems _______________ 26 Figure 3.2. The example of taxonomy of recommender systems ____________________ 28 Figure 3.3. The example of taxonomy of the recommender systems _________________ 28 Figure 3.4. The taxonomy of the recommender filtering and matching techniques _____ 30 Figure 3.5. The example of taxonomy of the recommender methods related to schefer __ 31 Figure 3.6. The classification of recommender systems presented in the master thesis __ 32 Figure 4.1. A simple weighted network _______________________________________ 35 Figure 5.1. Get result table from database _____________________________________ 38 Figure 5.2. Comparison between results ______________________________________ 38 Figure 5.3. The example of mutual friends and make suggestion ___________________ 39 Figure 5.4. The example of mutual friends and don’t make suggestion ______________ 39 Figure 5.5. Algorithm diagram _____________________________________________ 39 Figure 5.6. Coding of sorted array descending of mutual friends comparison _________ 40 Figure 5.7. Example sorted array descending of mutual friend’s comparison _________ 40 Figure 5.8. Chart result of user1 mutual friends percentage _______________________ 41
LIST OF TABLES
Table 2.1. Show the Top 15 website and social networks by Alexa Rank in 2016 ______ 14 Table 3.1. The division of recommendation techniques ___________________________ 29 Table 5.1. Result of user (Accuracy) _________________________________________ 42 Table 5.2. Result of of JC and CN and PA metrics for user1 and all other user ________ 43
ABSTRACT
Friend Recommendation System in Online Social Networks
Social media provide an important source of information regarding users and their interactions which is very valuable for the recommender systems. In web-based social networks social trust relationships between users indicate the similarity of their needs and opinions. In our paper, we presented a social network based recommender system app that utilizes the information of user and makes recommendations by considering users weight however we measured how many mutual friend they have been make suggesting and calculating the weight between each user by same formula and rule, then make recommending friends. We also help the users in a way by searching and recommending friends who do not belong to the same category of the major interest as the user but they have many mutual friends but they are not friends. Although there has been much work done in the industry and academia on developing the theory and application of social networks as well as recommender systems, the relation between these research areas is still unclear. An innovative idea, which enables to integrate these areas, and applies recommendation systems to the online social network systems, is proposed in this thesis. Recommendation systems for social networks differ from the typical kinds of recommendation solutions, since they suggest human beings to other ones rather than inanimate goods. Thus, conventional recommendation methods should be enhanced by social features of the networks and their members.
Keywords: Social media, Recommender system, Online Social Network Systems.
ÖZET
Çevrimiçi Sosyal Ağlarda Arkadaş Tavsiye Sistemi
Sosyal ağlar, tavsiye sistemleri için kullanıcılar ve onların ilişkileri ile ilgili bilgiler sağlayan çok değerli bir bilgi kaynağıdır. Web tabanlı sosyal ağlarda, kullanıcılar arasındaki güçlü benzerlikle ve ilişkiler onların benzerliklerini göstermektedir. Çalışmamızda kullanıcıların benzerliği ve ihtiyaçları bakımından sosyal ağ tabanlı bir ağırlıklandırılmış tavsiye sistemi önerilmiştir. Aynı zamanda çalışmamız da kullanıcının diğer bir kullanıcı ile ortak arkadaş sayısına bakarak aradaki bağın kuvveti bulunarak yeni arkadaş önerimi yapılmıştır. Bununla birlikte sistemimiz kullanıcıların kendileri ile aynı grupta bulunmayan veya benzer özelliklere sahip olmayan ama ilişkili olduğu kullanıcıları arama yaparken hızlı bulmasına da yardımcı olmaktadır. Aslında bu alanda hem endistüriyel hem akademik çok çalışma yapılmış olsa da halen gelişmeye açık bir alandır. Bu tezde tavsiye sistemlerinin çevirimiçi sosyal ağlarda kullanımı konusunda yenilikçi bir fikir sunulmuştur. Sosyal ağlarda tavsiye sistemleri diğer temel tavsiye sistemlerinden ciddi oranda farklıdır, çünkü çevirimiçi sosyal ağlar canlıdır insanların hareketlerine göre değişmektedir. Bundan dolayı geleneksel metodlardan öte sosyal ağda bulunan bağlantılar ve ağı kullanan kullanıcılar arası ilişkiler hesaba katılmıştır.
Anahtar Kelimeler: Sosyal medya, Tavsiyeci sistemi, Çevirimiçi Sosyal Ağ Sistemleri.
1. INTRODUCTION
So as to beat data additional heap, recommender systems have turned into a key apparatus for furnishing clients with customized suggestions on components, for example, music, books, news, and pages [1-3]. Recommender systems (RS) are ones used to prescribe clients to clients rely on upon a few criteria. They are typically used to handle and understand the data over-burden [4, 5].
The late rise of online social networks gives us gigantic measure of data identified with client conduct and companion connections have shown their significance to build up a productive recommender framework in this field. Facebook's companion recommender framework is relying upon the idea of social diagrams. It highlights clients as the clients you may know through associations on the client's profile, i.e., shared companions established on their work, training points of interest, systems and so forth.
For instance, if both two client's C and D include a similar school in their instruction, then both C and D will be appeared with the client they may know on each other's profile or if both C and D have substantial number of common companions then Facebook surmises that C and D may know each other thus the suggestion takes after. the recommendation depends on different think, first method may be make suggesting when two user have same
location or position or have like to same page or group, the second method depend on mutual friend they have and social media like Facebook predict the link between this two user C and D by measuring the weight (the number of mutual friend between C and D) then make suggestion. in this paper we will explain the second method how to implemented with the social media like Facebook [1,6 - 8].
The idea of both online social networks and recommender systems was developed by many researches. All existing recommendation methods suggest products or services to people [9 - 11]. Whereas method presented in this thesis recommends one human being to another. The whole idea of applying recommender systems in social network is innovative, thus many new challenges have occurred. Building such a framework requires integrating two separate branches of knowledge: computer science and sociology. This makes work even more interesting because enables to create interdisciplinary groups in which people with different backgrounds can cooperate with each other’s. On one hand, when we recommend one person to another, we should have deep knowledge about algorithms that can be utilized and pick the appropriate one. Usually, the chosen method must be further tailored to the specific needs of the system. Additionally, it is common practice that the creators of RS integrate many different recommendation methods because some techniques complement other ones. In other words, one technique usually copes with the shortcoming of other one and this relationship is bidirectional. Nevertheless, not only the knowledge about recommender systems is required. On the other hand, we should know a social behavior of people who create the network and be aware of their preferences.
In the recommender system for online social network not only the needs and preferences of a person who expects some suggestions must be considered, but also the expectations of a person who will be suggested. This factor makes the whole problem even more complicated but also more fascinating [12].
1.1. Related Work
Sandeep Konjere, V. N. Dhawas,[13], explained how to associated with world online correspondence and sharing data utilizing the interpersonal organization destinations turn into an extremely celebrated as of late. In any case, it's extremely testing work for informal community site to give the protection and security. Be that as it may, clients need to end up
new companions to build their social associations and in addition to get data from particular gathering of individuals. Numerous online informal communities (OSNs) utilizing the past Friend suggestion strategy and which is exceptionally prominent now days. There is a tremendous necessity to actualize security protecting companion suggestion strategies for web-based social networking as client security is primary rationale these days. Online Social Networks(OSNs), are not just get the centered from millions individuals to invest their energy consistently on Social network additionally staggeringly actualize OSN customer’s groups of friends by utilizing sidekick recommendations. This paper is spurred by need of companion proposition without indicating protection and security when utilizing informal organizations. The objective in executing of our framework is to bolster clients of OSN by safely makes trust with a odder which is achieving by multi-jump proposal prepare. Dynamic OSN client security assurance by utilizing proposed technique which is permitting them to improved their Social network. To do secured social coordinated organizing, existing system utilize the safe kNN arrange. However, with the assistance of KNN, separation based learning is not clear which kind of separation to utilize and their part to use to give the best results and estimation cost is high. To overcome on this impediment and expanded the results exactness, proposed system uses SVM classifier for secure social arrange coordinating. Through security investigation and trial results, we exhibit that the security, feasibility and precision of the proposed method is to have superior to anything existing one.
Katarzyna Musiał, [14], although there has been much work done in industry and academia on developing the theory and application of social media as well as recommender systems, the relation between these research areas is still vague. An innovative idea, which enables to integrate these areas, and applies recommendation method to online social mediasystems, is proposed in this thesis. Recommendation systems for social media differ from typical kinds of recommendation solutions, since they propose human beings to other ones rather than inanimate goods. Thus, conventional recommendation system must be improved by social features of networks and their members. This thesis presents the outcome of the study on recommendation system framework for virtual communities. It also contains an overview of recent approaches to recommendation method and social media, as well as description of online social network systems.
Ph. Eijlander, [15], in this postulation, they research how recommender frameworks can be connected to the space of social bookmarking. All the more particularly, they need to
examine the errand of thing suggestion. For this reason, intriguing and applicable things— bookmarks or logical articles—are recovered and prescribed to the client. Proposals can be founded on an assortment of data sources about client and things. It is a troublesome assignment as we are attempting to foresee which things out of a vast pool would be appropriate given a client's advantages, as spoke to by the things which the client has included the past. In our investigations we recognize two sorts of data sources. The first is to use information incorporate into the folksonomy, which speaks to the past determinations and exchanges of all clients, i.e., who included which things, with what labels. The second data source is the metadata depicting the bookmarks or articles on an online networking bookmarking, for example, title, portrayal, creation, labels, and fleeting and production related metadata. they are among the first to explore this substance based part of suggestion for social bookmarking sites. they analyze and consolidate the substance based favor the more normal use based methodologies. On account of the oddity of applying recommender techniques to online networking bookmarking, there is not an expansive assortment of related work, results, and outline standards to build on. They in this manner adopt a framework based strategy for the assessment our work. They attempt to reproduce, as sensibly as could be expected under the circumstances, the reaction of the client to various variations of suggestion calculations in a controlled research facility setting. They concentrate on two particular spaces: (1) suggesting bookmarks of Web pages and (2) prescribing bookmarked references to logical articles. It is essential to comment, nonetheless, that a framework based assessment can just give them a temporary gauge of how well our calculations are getting along. Client fulfillment is affected by more than just proposal precision (Her locker et al., 2004) and it is basic to catch up our work with an assessment on genuine clients in sensible circumstances. Notwithstanding, this is not the concentration of the postulation, nor they will concentrate on undertakings, for example, label proposal or discovering similarly invested clients. they concentrate entirely on recommending items.
Alan E. Mislove, [16], clarified the online web-based social networking locales have detonated in fame. Various locales are committed to find and save gather and to find and share various types of substance. Online web-based social networking speak to another kind of data system that varies fundamentally from existing systems like Web. For instance, in the Web, hyperlinks between substances frame a chart that is used to sort out, explore, and
rank data. The properties of Web diagram have been adapted broadly, and have prompted to valuable calculations, for example, PageRank. Conversely, few connections exist between substances in online web-based social networking and rather, the connections exist amongst substance and people groups, and between people groups themselves. Be that as it may, few is known in the examination group about the properties of online web-based social networking charts at scale, the components that shape their structure, or the ways they can be utilized in data techniques. In this postulation, they utilize novel estimation strategies to contemplate online web-based social networking at scale, and utilize the subsequent experiences to outline creative new data techniques. Initially, they look at the structure and development examples of online web-based social networking, concentrating on how companions are associating with each other. They lead the primary expansive scale estimation investigation of different online web-based social networking at scale, catching data about more than 50 million companions and 400 million connections. they examination recognizes a typical structure over various media, describes the fundamental procedures that are embellishment the system structure, and uncovered the rich group structure. Second, we influence our comprehension of the properties of online web-based social networking to outline new data techniques. In particular, they fabricate two unmistakable applications that influence diverse properties of online web-based social networking. They display and assess Ostara, a novel framework for forestalling undesirable correspondence that influences trouble in building up and keeping up connections in online networking. They additionally present, convey, and assess Perspective, a framework for improving Web look utilizing the normal group structure as a part of online networking. Each of these frameworks has been assessed on information from genuine online web-based social networking or in a sending with genuine clients.
Jilin Chen, [17], explained Interpersonal organization locales have encountered a blast in both quantity of clients and measure of client contributed content in a years ago. Today, a large number of individuals utilized Facebook, Twitter and so forth to keep companions up, to take part in arbitrary prattle, and to share and devour pictures, news, valuable tips and fun stories. Numerous dynamic companions of online networking destinations are, be that as it may, continually harried by data over-burden – there are an excessive number of different companions to connect with and a lot of substance to peruse. Therefore, trouble in finding right companions and substance to concentrate on has been
recognized as a key challenge for online networking destinations. In this proposal they ask for to meet the resist by planning three customized recommender media. The main framework is a companion recommender, which helps people groups discover potential companions to associate with via web-based networking media locales; the second framework is a data recommender, which helps people groups find intriguing bits of data from their online web-based social networking; the third is a discussion recommender framework, which helps people groups find fascinating discussions happening around their online web-based social networking. In planning recommender strategies, we adjust calculations from related areas and make new calculations. They refine these calculations through disconnected information examination. They then plan and enhance individuals interfaces of recommender strategies through participatory outline. With the full recommender strategies outlined and executed, they convey these techniques to genuine clients of web-based social networking locales, and assess an assortment of recommender calculations online through client ponders.
1.2. Organization of The Thesis
Chapter 2: Social network brings a short definition of network, information about social networks, types of social networks and social networks analysis. The past researches related to social network are addressed in this chapter.
Chapter 3: Recommendation method starts with the type of recommendation method. The past researches related to recommendation method are addressed in this chapter.
Chapter 4: Link Prediction for Recommendation in social networks a weighted system is a system where the ties among hubs have weights allotted to them.
Chapter 5: Proposal method in (OSN) introduce a recommendation method for predicting new way to recommend friends in online social networks. It is shown with experimental results at the end of this chapter.
Chapter 6: Discussion and Conclusion summarizes and discusses contributions of this research work. It also discusses limitations of features introduced in this research work, and future directions of our research
2. SOCIAL NETWORKS
The continuously evolving relationships between people and create the greater need of communication. This necessity makes not only the development of the online social network systems (OSNS) like Friendster or LinkedIn [18], but it also increases the popularity of typical communication services like instant messengers, web conferences, Voice over IP systems (VoIP), etc. Online social media support the interpersonal communication and creation as well as evolution of human communities and because of that they are in the area of interest both sociologists and software engineers.
2.1. Social Networks
The social networks (SN) have recently become more and more important element of information society. The social network problem is very wide and its concept is developed in many research areas such as corporate partnership networks [19], scientist collaboration networks [20], film actor’s networks [21], etc. The huge variety and many possible areas where social media can be applied cause that they have become subject of many researches. Since the relationships from social network, their maintenance and quality reflect social behaviors of individuals, the research on networks can be helpful at qualitative assessment and quantitative of human relationships in the age of information community. The concept of SN is used to describe the relationships between friends, co–workers, members of particular community, relatives in the family. Not only character of the relationships can be analyzed, but also their force and direction. Although social network analysis emphasizes connections between users, the results of SNA provide also much information about individuals themselves. There are many, different social media and the taxonomy of social media is not established.
Sociologists have considered a significant number of properties of disconnected Social network, and we just quickly depict a couple of applicable discoveries. For a more entire outline of disconnected informal communities and related examination systems, we allude the peruse to the book by Wasserman [22]. Milgram [20] demonstrated that the normal way length between two Americans was six jumps, exhibiting that online networking can be named little world.
Pool and Kochen [23] gave an investigation of how the little world property of informal organizations influences contacts and impact. The powerful paper by Granovetter [24] contended that an informal organization can be parceled into "solid" and "feeble" ties, and that the solid ties are firmly grouped, while the frail ties speak to longer-separate connections.
We could confirm that online informal organizations have comparable properties, with short way lengths and solid groups associated by long separation joins. As online Social network picked up notoriety, scientists have started to research their properties. Adamic et al. [25] concentrated on an early online interpersonal organization at Stanford University, and found that the system has little world attributes and also a huge grouping coefficient. Liben-Nowell et al. [26] found a solid connection amongst's kinship and the geographic area of clients by utilizing information from Live-Journal. Kumar et al. [27] inspected two online informal organizations from Yahoo! also, found that both had a prevailing SCC. Girvan and Newman watched that clients in online Social network tend to frame firmly sew bunches [28], prove by a high grouping coefficient.
We could check these properties on various destinations and on a much bigger scale in our study. In later work, Ahn et al. [29] broke down total information from the substantial South Korean long range informal communication site Cyworld [30], alongside information from little example creeps of MySpace and Orkut. The creators got information specifically from CyWorld administrators, and the volume of accessible information lets the creators to lead a top to bottom investigation of that site utilizing a portion of similar measurements that we use in this proposition. The examination with various systems, then again, is constrained by the little crept information example of MySpace and Orkut. Our study is to a great extent integral: the information accessible to us for any one site is less nitty gritty, yet we can look at huge slithered information sets from numerous locales. At last, analysts have likewise analyzed how the movement organize, or the example of associations between clients, contrasts and the informal community. Specifically, Wilson et al. [31] examined the action system of tests of the Facebook system and found that, rather than the interpersonal organization, the movement system is much sparser and has an altogether bring down maximal degree.
Chun et al. [32] discovered comparable properties for the connection organize in CyWorld. In our work, we concentrate just on the informal organization, however our
approach and techniques could be normally connected to the movement arrange also. Social networks can be partitioned into a few gatherings as far as various standard. Online networking can be: committed (e.g. dating or business systems, systems of clients, graduates, fun clubs), aberrant (online communicators, address books, messages) regular exercises (e.g co–authors of logical papers and co–organizers of occasions), neighborhood systems (e.g. individuals living in the area), families, employee’s media, hyperlink media (links between home pages), etc [33]. To put these different kinds of networks in order, they can be classified based on the type of the relationship that connects two persons [33]. In this case, business and social connections can be distinguished (Figure 2.1). The former ones contain social networks that consist of people who are joined to each other because of things they do together but simultaneously they do not share their special lives. Those can be called professional networks. Company – the employees create the social network of co–workers can be a good example. Likewise, individuals who sort out together e.g. a meeting or other occasion, make interpersonal organization of co–organizers. These individuals are associated in light of the fact that they cooperate and their participation more often than not brings some result, e.g. article, meeting, book, and so forth [33].
Figure 2.1 The division of social networks based on the type of the relationship
Moreover, the grouping of web-based social networking can be construct not just with respect to the kind of relations that happen in system, additionally on the sort of the correspondence channel between individuals that serves to exchange resources i.e. they can be either in person or device supported (virtual, via computer, phone, snail mail, etc.) (Figure 2.2) [33].
Figure 2.2 The division of social networks based on type of communication channel
2.2. Online Social Networks
Characterize an online web-based social networking to be a framework where (a) clients are top of the line substances with a semi-open profile, (b) individuals can make unequivocal connections to different people groups or substance things, and (c) people groups can explore informal organization by perusing connections and profiles of different people groups. This definition is reliable with that utility in past studies [34]. Online webbased social networking fill various needs, however over all destinations three essential parts emerge as basic. In the first place, online social network are utility to keep up and reinforce existing social ties, or make new informal organizations associations. The locales permit people groups to "well-spoken and make noticeable their web-based social networking", there by "speaking with people groups who are as of now a piece of their developed interpersonal organization" [34]. Second, every client is utilized online web-based social networking to transfer her own particular substance. The substance shared frequently fluctuates from site to site, and some of the time is only people groups profile itself. Third, online web-based social networking are utilized to discover new, fascinating substance by separating, prescribing and sorting out the substance transferred by peoples. Online social networks provide separate and simple techniques to correspond with each other’s users and obtain new friends on the social media. Badly, privacy of friend and their data concerns brought up in recommendation process obstruct the development of OSN clients companion circle [35]. Some OSN clients want to secure their private information from the unaware peoples and show their companions’ data to the general domain [35]. To overcome on this
issue, utility privacy-preserving trust-based friend recommendation scheme for online social media, which boost two outsiders design a trust connection in light of the multi-hop trust chain. For classification existing system uses KNN algorithm [35].
Online social media can be defined as the set of people who are linked by a computer network [36]. Online social media enable and facilitate not only to form new connections but also the communication between users who are in better places and on various timetables [37]. This makes connections not as unmistakable as those from this present reality. However, it can turn into a big advantage, when people cannot meet with each other but they can communicate in other way (e.g. sending e–mails). Then again, in online social network people lose the opportunity to study some of the verbal (e.g. voice tone) and nonverbal (e.g. body language) elements of the conversation [37]. This relative lack of social presence allows developing the community organized by shared interest not by shared neighborhood [38] According to one of the divisions of social networks presented above, (Figure 2.2) they can be called PC upheld Social network (CSSN). In this group two subsets can be distinguished: non ongoing and real time online social networks (Figure 2.3).
The former ones enable the offbeat correspondence between two people or from one individual to a gathering of individuals [37]. The illustration here can be electronic mail. When person x sends an e–mail to person y, the relationship between these people comes to existence. However, if person x only has e–mail address of person y but they have never sent e–mail to each other, then the bound between them does not exist. The internet forums, in contrary to e–mail that support the communication between two persons or small group of people, enables all people from particular social network to read all messages that were submitted by every single member of network. Their functionality is similar to bulletin board from real world.
2.3. Functions of Online Social Network Systems
In order to introduce some systematic, the distinction between Online Social Networks and Online Social Network Systems should be done. The former ones are all social networks that are supported by computers and communication channel in this case is computer network [36]. For example, the employees from one company who work all over the world create the social network, in which sending e–mails, using instant messengers, and VoIP systems recompense them for the lack of in person communication [33]. The latter ones can be also called virtual communities and their main goal of these systems is to create and maintain the social networks. These systems utilize such tools as: e–mails, chats, etc. Nevertheless, online social networks systems are the subset of the online social networks, because their existence is possible only when the members have access to computers and computer networks. Before defining any functionality of OSNS, the data that are available in such networks should be identified [33]. All OSNS are characterized by the static traits of performing artists like their advantage or demographic information and also the depiction of the association between on-screen characters. The greater part of this data makes people groups profile. By and by, not just the average data about the specific client like their advantage, demographic information, and so on can be considered, additionally their exercises and a few measures of association with different people groups, particularly those identified with the procedure of introduction of limits and some further, consecutive exercises [33].
The social articulation of the client x in the system comprises of two information sets: general and accumulated openness and exercises components of this people groups in connection to all others, likewise before, and measures of relationship between client x and different individuals from the system [33].
The most significant elements in the first set are the user's willingness for initialization of relationships and reply to invitations from others. The most important case within the frequency and intensity are second set with which all relationships are maintained, called also common communication [33].
2.4. Features of Social Networks
Each social network can be characterized by a set of features that provide information about the overall structure of the network, e.g. range [36], density, reachability, and connectivity [39] as well as about each of the actors, e.g. centrality and roles [36]. Range of the social network can be assessed by such characteristics as size [39] and heterogeneity (also called composition) [37]. Small, homogeneous groups are good for conserving existing resources whereas large and heterogeneous networks are good for gathering and exchanging new resources [36]. Traditional work groups, families, town communities are examples of the former. To the latter belong e.g. online communities.
Virtual communities belong to most heterogeneous groups, because they connect people with totally different cultural and religious backgrounds. Density is ratio of the nodes that exist within network to all possible nodes [39]. When density of network is known, the speed with which information diffuses among the actors and “the extent to which actors have top levels of social capital and/or social constraint [39] can be defined. One of network features is reachability that defines actor x as reachable by actor y when there is a set of connections which enables to trace from x to y [39].
Additionally, it is not important how many people fall between x and y. Reachability pinpoints if there is connection between two actors, whereas connectivity shows in how many possible ways user x can reach y. If many different paths that connect two actors exist, then they have high “connectivity” [39]. Centrality is feature that provides precise description of network structure. It serves to identify individuals that are central or isolated nodes of network [36]. Many measurements to centrality problem exist, e.g. Freeman’s or Bonacich’s approaches indicate to degree centrality [39]. People who maintain the biggest number of connections are called members with “high degree” [36] and they may have in advantaged positions. [39]. Some researches distinguish “cut–points” [39]. This term refers to members whose departure from the network would cause its disintegration. Degree as well as closeness and betweenness are elements that decide about member’s power in network [39]. Closeness centrality can be defined as the length of the paths from one actor to another. Shorter path means that actor has better position and his power within the network is bigger [39]. If the member of the network is between other pair of people, then the power of such member increases and this feature is called betweenness centrality [39].
One of the features that also describe SN are roles which can be assigned to members according to their behaviors. Similarities in peoples’ behavior enable to distinguish for each network the set of roles [36]. These roles can be identified empirical by finding the regularities in the patterns of relationships (known as structural equivalence) [36]. To understand how a whole network is likely to behave the analysis of groups and sub structures should be done. The phenomena that can be observed within networks are cliques as well as strong and weak ties.
2.5. Top 10 Social Networks
Here are main 10 Most Popular Social Networking Sites as got from Alexa rank which is a ceaselessly upgraded normal of every site's Alexa Global Traffic Rankand U.S. Activity Rank from both Compete and Quant cast. Indicates a gauge for locales with constrained information.
Table 2.1 show the Top 15 website and social networks by Alexa Rank in 2016 1- Google 4- Baidu 7- Amazon 10- Live 13- Linkedin 2- Youtube 5- Yahoo 8- Qq 11- Taobao 14- Instagram 3- Facebook 6- Wikipedia 9- Twitter 12- Vk 15- Bing
Facebook Founded: 2004.
Revenue: US$800 million. Alexa Rank: 3. Twitter Founded: 2006. Revenue: US $150 million. Alexa Rank: 10.
Founded: 2002.
Revenue: US$161.4 million . Alexa Rank: 13.
Figure 2.4 Ranked number of active users in million (2016)
2.6. Social Network Analysis
Researches on social network analysis (SNA) stretches back more than half a century [40] where Jacob L. Moreno often is credited to be the researcher who was first systematically make utilize of social network analysis like techniques [41].
To characterize Social network analysis, it concentrates on structure of connections, extending from unplanned associate to close securities. Interpersonal organization
Formal and casual connections It maps and measures to comprehend what encourages or blocks the learning streams that quandary interfacing units, viz., who knows whom, who work with whom and who offers what data and information with whom by what correspondence systems (e.g., information and data, voice, or video communications) [42]. Social system investigation is a strategy with developing application in sociologies and has been connected in various region like connection expectation from social network, brain science, wellbeing, business association, and electronic interchanges. Social network analysis is considered in a vast amount of areas. To mention just a few, it can be used for understanding social interactions, to optimize flow of information between employees in a company, or to study and analyze criminal or terrorist organizations. Important problems within social media analysis are, among others:
• To collect and extract useful data
• To visualize in a way that support analysts with interpretation of social structures. • To identify important structural patterns of the network (such as the identification of
actors in network that are extra important or powerful). • To predicate new links by calculating closeness, centrality.
Analyzing social networks empowers us to distinguish a few entomb and intra associations between hubs in and outside their systems. The examination of most social network is a clarification of ties part of every hub in the system [43]. The qualities clarifying the part of hubs can be measured as centrality qualities. There are 4 classifications of centrality values to be specific degree centrality, closeness centrality, betweenness centrality, eigenvector centrality.
2.6.1. Basic Metrics in Social Network Analysis 2.6.1.1. Degree Centrality
Degree centrality is characterized as the quantity of connections accident upon a hub (i.e., the quantity of ties that a hub has). In the coordinated system (implying that ties have course), we more often than not characterize two separate measures of degree centrality, to be specific in degree and out degree. In degree is a check of the quantity of binds coordinated to the hub, and out degree is the quantity of ties that the hub coordinates to others.
Formally for a graph 𝐺 ∈ (𝑉, 𝐸) with n vertices, the degree centrality C´D (vi) for
vertex vi is as follow:
C´D (vi) = (2.1)
Figure 2.5 An example network for degree centrality calculation
C´D (4) =5/7 =0.71 C´D (7) =2/7 = 0.28
In the figure 2.5, node 4 has the highest degree centrality while 7 and 8 have the lowest.
2.6.1.2. Closeness Centrality
Closeness centrality is the total number of spaces which an object has from other objects in the network, it shows centralization of the under analysis object. Intuitively we say two sets are close if they are arbitrarily close to each other.
In diagram hypothesis closeness is a centrality measure of a vertex inside a shape. Vertices that are "shallow" to different vertices (that is, those that have a tendency to have short geodesic separations to different vertices with fit as a fiddle) have higher closeness. In the system hypothesis, closeness is an advanced measure of centrality It is knowing as the mean geodesic separation (i.e., the most limited way) between a vertex v and all different vertices reachable from it.
Closeness centrality is entirety of all separations from a fascinating hub to different hubs in the system. It clarifies whether such a hub is the focal point of the system or not. Formally for a graph 𝐺 ∈ (𝑉, 𝐸) with n vertices, the closeness centrality CC (vi) for vertex
CC (2.2)
Figure 2.6 An example network for closeness centrality calculation
CC (4) = 8-1/1+1+1+1+1+2+2 = 7/9 = 0.77 CC (5) =8-1/1+1+1+2+2+2+3 =7/12
=0.58
In figure 2.6, node 4 is more central than 5.
2.6.1.3. Betweenness Centrality
Betweenness is a centrality measurements of a hub inside a shape (there is likewise edge betweenness, which is not talked about here). Vertices that happen on numerous most limited courses between different vertices have higher betweenness than those that don't. For a diagram G∈(V,E) with n vertices, the betweenness CB(vi) for vertex vi is figured as takes after:
• For every match of vertices (s,t), process all most limited courses between them. • For every combine of vertices (s,t), decide the portion of most limited ways that go
through the vertex being referred to (here, vertex vi).
• Whole this portion over all sets of vertices (s,t). On the other hand, more compactly:
(2,3) In the equation (2.3) 𝜎𝑠𝑡(𝑣𝑖) is the number of shortest paths from s to t, and 𝜎𝑠𝑡 is the number of shortest paths from s to t that pass through a vertex vi.
2.6.1.4. Eigenvector Centrality
Eigenvector centrality is a measure of the significance of a hub in interpersonal organization. It doles out relative scores to all hubs in the system in light of the rule that associations with high scoring hubs contribute more to the score of the hub being referred to than equivalent associations with a low scoring hubs. Google's PageRank is a variation of the Eigenvector centrality measure.
2.6.1.5. Other Metrics Bridge:
Bridge is a link; if we delete it then the nodes which formed the mentioned link lie on different subgraphs.
Centralization:
Centralization is the contrast between the quantities of connections for every hub separated by greatest conceivable total of contrasts. A unified system will have a considerable lot of its connections scattered around one or a couple of hubs, while a decentralized system is one in which there is a couple of varieties between the quantities of connections every hub has. Clustering Coefficient:
A measure of the likelihood that two partners of a hub are partners themselves. A higher bunching coefficient shows a more prominent 'cliquishness'.
Cohesion:
Union is how much on-screen characters are associated specifically to each other by firm bonds. Gatherings are indicated as "coteries" if each individual is straightforwardly fixing to each other individual, 'groups of friends' if there is low stringency of direct contact, which is uncertain, or as basically durable squares if exactness is needed [44].
Degree:
Degree is the sum of the links that an object/node has in a network. Definitely it is the sum of in-degree and out-degree of a node.
Density:
The degree a respondent's ties know each other/extent of ties among a people chosen people. Arrange or worldwide level thickness is the extent of ties in a system in respect to the all number conceivable (scanty versus thick systems).
Local Bridge:
Nearby extension is an edge is a neighborhood connect if its endpoints share no normal neighbors. different a scaffold, a nearby extension is contained in a cycle.
Path Length:
Way length is the separations between couples of hubs in the system. Normal way length is the normal of these separations between all couples of hubs.
Prestige:
Path length is the distances between couples of nodes in the network. Average pathlength is the average of these distances between all couples of nodes.
2.7. Related Work
Gowsalya, Kumar Pravin, [45], clarified from twenty years back, individuals ordinarily made companions with other people who live or work near themselves, for example, neighbors or associates. With the quick advances in informal organization administrations, for example, Facebook and Twitter and Google+ have given us progressive approach to making new companions. In this paper, which prescribe the companions to their client in view of their way of life rather than social diagram. By utilizing sensor rich advanced cell and afterward roused with the content mining, we display everyday life of client as an existence archive, from this the ways of life and sorts are extricated by Latent Dirichlet Designation Calculation to promote utilize the closeness metric to gauge the comparability of way of life between the client and ascertain the client affect with the companion coordinating diagram. At long last, the framework to incorporates the input system to enhance the exactness of suggestion.
Chaney, Allison J.B., BLEI. David M., [46], demonstrated preference-based suggestion frameworks have changed how they devour media. By examining use information, these ways reveal our inactive inclinations for things, (for example, articles or motion pictures) and shape suggestions in light of the lead of others with comparative tastes. In any case, conventional inclination based suggestions don't represent the social side of
utilization, where a trusted companion may guide them toward an intriguing thing that does not coordinate them run of the mill inclinations. In this work, we plan to cross over any barrier amongst inclination and social-based suggestions. They create social Poisson factorization (SPF), a probabilistic model that consolidates online networking data into a conventional factorization way SPF acquaints social viewpoint with algorithmic proposal. they build up a versatile calculation for examining information with SPF and imagine that it outflanks contending strategies on six genuine datasets information sources incorporate a social peruser and Etsy.
Seth Aaditeshwar, and, Zhang Jie. [47], clarified given quick development of participatory media substance, for example, online journals, there is a need for configuration customized recommender technique to prescribe just helpful substance to clients. They trust that notwithstanding delivering helpful suggestions, affirmed bits of knowledge from media research, for example, disentanglement and assessment differing qualities in proposals ought to frame the premise of such recommender technique, so that the conduct of the strategies can be seen all the more nearly, and adjusted if essential. They propose and assess such a framework in view of a Bayesian client show. They utilize the hidden web-based social networking of blog writers and perusers to show the inclination highlights for individual people groups. The underlying consequences of our proposed arrangement are empowering, and set motivation for future research.
Yang Xiwang , Steck Harald [48], this theory demonstrates that online web-based social networking data guarantees to build suggestion accuracy past the abilities of absolutely appraising input driven recommender frameworks (RS). As to better serve client's exercises crosswise over deferent spaces, numerous online web-based social networking now bolster another element of clients Circles, which refines the area unmindful Friends idea. Recommender framework ought to likewise profit by area particular Trust Circles. Naturally, a client may trust deferent subsets of companions with respect to deferent areas. Shockingly, in most existing multi-classification rating datasets, a companion's social associations from all classes are combined. This paper shows a push to create circle-based RS. They concentrate on construing classification particular social trust hovers from accessible rating information joined with interpersonal organization information. They out-line a few variations of weighting companions inside circles in light of their derived aptitude levels. Through examinations on clearly accessible information, they show that the proposed
circle-based suggestion models can better use companions social trust data, bringing about expanded proposal exactness. Gartrell Charles Michael, [49], this proposition demonstrates that joining social pointers enhances the prescient execution of gathering and individual in view of recommender techniques. They dissect the effect of social pointers through little and expansive scale studies, execute and assess new suggestion models that join our bits of knowledge, and imagine the practicality of utilizing these social markers and other relevant information in a sent versatile application that supply eatery proposals to little gatherings of clients.
Naruchitparames Jeff, Hadi G¨Unes Mehmet, [50], clarified Social media destinations utilize proposal strategies in commitment to giving better client encounters. The intricacy in creating proposal strategies is generally because of the heterogeneous way of web-based social networking. This paper introduces a way to deal with companion proposal techniques by utilizing complex system hypothesis, psychological hypothesis and ideal hereditary calculation in a two-stage way to deal with give quality, client suggestions while all the while deciding an individual's impression of fellowship. Our exploration accentuates that by joining system and hereditary calculations, better suggestions can be accomplished contrasted with every individual partner. They test them approach on 1,200 Facebook companions in which they watch the consolidated technique to beat absolutely social or simply arrange based methodologies. They preparatory results speak to solid potential for creating join proposal strategies utilizing this consolidated approach of individual interests and the fundamental system.
Cen Lei, et al, [51], clarified the fame of online web-based social networking is on consistent ascent because of various favorable circumstances, including on the web correspondence and sharing data of enthusiasm among clients. It is regularly that clients need to make new companions to grow their social associations and in addition to get data from an expansive scope of clients. Companion proposal is a critical application in numerous OSNs and has been concentrated broadly in the most recent past. Nonetheless, with the developing worries about client security, there is a solid need to create protection saving companion proposal frameworks for online networking. In this paper, they propose two novel strategies to suggest companions for a given companion by utilizing the regular neighbor's closeness measure in a security saving way. The main framework depends on the properties of an added substance homomorphic encryption conspire furthermore uses a
widespread hash work for proficiency point. The second strategy uses the idea of securing the source protection through unknown message steering and suggests clients precisely and productively. Moreover, we exactly think about the proficiency and exactness of the proposed conventions, and address the usage subtle elements of the two techniques by and by. The proposed conventions supply an exchange off among security, exactness, and effectiveness; in this manner, clients or the system provider can pick between these two conventions relying upon the hidden necessities.
Samanthula Bharath Kumar [52], clarified Online interpersonal organizations, for example, Facebook and Google+, have been developing as another correspondence benefit for clients to remain in touch and impart data to relatives and companions over the Internet. Since the clients are producing tremendous measures of information on interpersonal organization destinations, an intriguing inquiry is the way to mine this huge measure of information to recover helpful data. Along this course, informal organization examination has developed as an essential device for some business knowledge applications, for example, distinguishing potential clients and advancing things in light of their interests. Specifically, since clients are regularly intrigued to make new companions, a companion suggestion application gives the medium to clients to grow his/her social associations and impart data important to more companions. Other than this, it likewise upgrades the advancement of the whole system structure. The current companion suggestion strategies use interpersonal organization structure as well as client profile data. Nonetheless, these strategies can never again be material if the security of clients is thought about. This work presents an arrangement of security safeguarding companion suggestion conventions in light of various existing comparability measurements in the writing. Quickly, contingent upon the basic likeness metric utilized, the proposed conventions ensure the protection of a client's close to home data, for example, companion records. These conventions are the first to make the companion suggestion prepare conceivable in protection improved long range informal communication situations. Additionally, this work considers the instance of outsourced interpersonal organizations, where clients' profile information are encoded and outsourced to outsider cloud suppliers who give long range interpersonal communication administrations to the clients. Under such a situation, this work proposes novel conventions for the cloud to do companion proposals in a security protecting way.
Alharthi Haifa, [53], explained due to the growth of online shopping and services, various types of products can be recommended to an individual. After reviewing the current methods for cross-domain recommendations, they believe that there is a need to make different kinds of recommendations by relying on a common base, and that it is better to rely on a goal customer’s information when building the base, because the customer is the one common element in all purchases. Therefore, they suggest a recommender method that develops a personality profile for each product, and represents items by a collected vector of personality features of people who have liked the items. To build personality profiles for items (IPPs) they investigate two ways. The first way is called average-based IPPs, that reflect the average which represents each item with five attributes Big Five Personality values of the friends who like it. The second way is named proportion-based, which consists of 15 attributes that aggregate the number of fans who have high and average and low Big Five values. The system functions like an item-based collaborative filtering recommender; that is, it recommends items similar to those the friend liked.
Zhepeng Li, et al. [54], Sheng explained link recommendation, which suggests links to connect currently unlinked friends, is a key functionality offered by major online social networks. Salient examples of link recommendation include “Friends or Friends You May Know” on Facebook and LinkedIn as well as “You May Know” on Google+. The main stakeholders of an online social media include friends (e.g., Facebook users) who use the network to socialize with other friends and an operator (e.g., Facebook Inc.) that establishes and operates network for its own benefit (e.g., revenue). Existing link recommendation systems recommend links that are likely to be established by users but overlook the benefit a recommended link may bring to an operator. To address this gap, we define the utility of recommending a join and formulate a new research problem the utility based join recommendation problem. They then propose a novel utility based join recommendation method that recommends links based on the value, cost, and linkage likelihood of a link, in contrast to existing join recommendation methods which focus solely on linkage likelihood. Specifically, they method models the dependency relationship between value and cost and linkage likelihood and utility-based link recommendation decision using a Bayesian network, predicts the likelihood of recommending a link with the Bayesian network, and recommends links with the highest likelihood. Using data obtained from a major U.S. online social network, they demonstrate significant performance enhancement achieved.
3. RECOMMENDATION METHODS
Recommendation systems belong to the research area that is intensively developed because it helps to achieve higher return on investment by e–commerce stores and provide higher quality of services [9]. It is necessary to build the overall view of the RS, their goals as well as their advantages and disadvantages, in order to understand the whole concept of RS for online social network.
When people purchase products, they have to make decisions which items to buy. They also have to decide which news to read in open access portal or which film to watch in multimedia store. Their choice often depends on the other users’ opinion, especially in the e–commerce. The oldest version of recommendation is “word–of–mouth” opinion [56]. This is the method that most people use when they decide to buy e.g. a new laundry machine. Hence, they ask their friends who are trustworthy, which laundry machine they suggest to buy and why should they pick this one not the other one. At the beginning the recommender systems did not create a separate research area and their roots can be traced back to the cognitive science, information retrieval, forecasting theories, approximation theory and management science [9]. However, the growth of the Internet and the rise of the e–commerce solutions caused development of the online recommender systems since the mid–1990s they have become an important research domain [9]. The reason for that was opportunity to share the opinion between a vast numbers of people who use the Internet. Group Lens is first known project in the recommendation area that is lead by John Riedl, Joseph Konstan, and Loren Terveen from the Minessota University is Group Lens Research Group [57]. The roots of this project can be traced back to the year 1992 when the main goal of the system was to explore automated collaborative filtering. After that, the collaborative technique was applied in filtering the information in Usenet news [10]. Ringo agent was one of the first applications that provided personalized music recommendations [59], which became available on July 1st 1994 [58]. In this method the users provide ratings of the music articles. Based on these opinions the user profile, which changes over time, is created.
The profile enables to create the recommendations by utilizing the social filtering method [59]. This method can be treated as the automation of “the word–of–mouth” recommendation [58]. The application that utilized concept of the Ringo system was
Firefly’s system. This technology was further developed by Yahoo and Barnes and noble who signed up to use it [56]. Finally, this method was used by book dealer Amazon.com that introduced the Book Matcher system. At the beginning the Book Matcher was used for book recommendations, but later on, the system has started to recommend other types of items,
using also other methods of recommendation [56]. there are still many challenges in this area. Each of existing recommendation methods suffers from some shortcomings. The goal of the current research is to cope with these disadvantages by combining many approaches. Together (hybrid approach) and create the recommender system.
3.1. Definition of Recommender Systems
There are numerous meanings of recommender frameworks. One of the first was introduced by Paul Resnick and Hal R. Varian in 1997. They guarantee that "in a run of the mill recommender framework, individuals give suggestions as sources of info, which the framework then totals and coordinates to proper beneficiaries" [60]. These frameworks are generally characterized as far as their usefulness as the frameworks or specialists that recommend the items to the clients who buy items on e–commerce destinations [11]. The recommender frameworks assist the purchaser to settle on the choice what with buying. The recommender frameworks can be arranged as a result of the level of personalization into non–personalized and customized techniques [61] (Figure 2.1).
Figure 3.1 Levels of the personalization in the recommender systems
The previous techniques don't consider the attributes and inclinations of the clients, though the last firmly relies on upon the client profile. Notwithstanding, some exploration guarantees that the recommender frameworks are just those ones, which deliver customized proposals [9, 62]. At the end of the day, the yield of these frameworks is the individualized suggestion that aides the single client to items or administrations that satisfy their specific needs. Accordingly, they adapt to data over-burden superior to the non–personalized
techniques and empower to discover and buy the right things from the extensive measure of conceivable decisions. The customized proposal is construct either with respect to the demographic data about clients or on the investigation of the past conduct of the client keeping in mind the end goal to anticipate their future conduct (communitarian and content– based separating) [46]. In addition, the personalization can be either constant or fleeting [61, 11]. Constant personalization, in view of the past clients' practices, empowers to make extraordinary rundown of items for every client. The necessity that should be satisfied in this circumstance is that clients must sign into the framework keeping in mind the end goal to make client profile for each of them. In a determined customized suggestion every individual on the Web website sees diverse proposals since they depend straightforwardly on the clients' close to home information. The proposals depend on the data got from the study reactions, acquiring history, items appraisals, and so forth. The client profile is a bit much in the fleeting personalization. For this situation the proposals are made by clients' practices amid a present session, their route and choice [11]. In this method the proposals are the same for all clients [61].
3.2. Goals of Recommender Systems
Recommender systems became an important and almost integral part of recent web sites; what is more, the vast number of them is applied to e–commerce. Jeff Bezos, CEO of Amazon.com, said: “If I had 3 million customers on the Web, I should have 3 million stores on the Web” [11]. Why do people believe that personalization and recommendations are a crucial part of e–commerce.
The aim of these systems is to help the potential buyers to pick the appropriate product to buy, so that they can be seen as decision support systems. On the other hand, they serve as the marketing help for the e–commerce stores because they increase the attractiveness of the offer. The main goals of the recommender systems are as follow:
• To adapt to data over-burden [9, 10, 59,63]
• To help all clients (new, regular, and occasional) to settle on choices what items to purchase, which news to peruse next [64], which motion picture merits viewing, and so on. • To change over eyewitnesses to purchasers
