Recommender System

1.      Introduction

In this information age, everyone is forced to make every decision, restaurant to go, movies to watch, or books to buy, from abundance of information. In those moments, people always turn to friends and others who are knowledgeable about our tastes and preferences for recommendations. Obtaining good recommendations becomes a vital issue when overwhelmed with decision making process.
Internet offers users accessibility to huge amounts of information, and consequently, providing good recommendations is one of the most pressing missions for the implementation of electronic environments. Therefore, the concept of Recommender Systems (RS) is introduced and studied. RS is focusing on assisting users to deal with information overload by recommending “attractive” items, based on users’ respective tastes and preferences, from huge databases or catalogues.

2.      Types of recommender system

There are three types of recommender systems, which are commonly recognized based on the adopted method of making item suggestions (Adomavicius & Tuzhilin, 2005): content-based filtering (CBF), collaborative filtering (CF), and hybrid filtering (HF).
1). CBF recommends the user items similar to the ones she preferred in the past. This approach is an efficient recommending method, only relay on the user itself, his/her preference and taste. However there are some intrinsic limitations, as it could be very difficult to extract the content for some items, and the recommending scale is very limited by the content based retrieval (same words).
2). CF recommends the user items that people (called neighbors in the literature) with similar tastes and preferences liked in the past. CF is adopted in numerous studies and researches, and is preferred by most researchers, but there are some limitations. First, a “cold start” problem, which occurs a new user is introduced, for whom there is no preference data. Moreover, the recommendation system choose neighbors only depends on the similarity between users and their neighbors, without considering global characteristics of the neighbors, which would make them less appropriate, such as their confidence or trustworthiness.
3). HF is a combination of content-based and collaborative filtering approaches..

3.      Recommender Agent

Recommender Agent (RA) is a crucial factor to a success online retailers, like Amazon.com and service providers like Netflix.com, as it provides a customized online shopping experience. It is speculated by many researchers that RAs make it possible for online merchants to guiding customers’ behavior (Gretzel & Fesenmaier, 2007). Felfernig and Gula demonstrated their view that RAs could convince the customer that some product attributes are more important than others or make the consumer more satisfied with their online shopping experience in Reference [3]. The theoretical model of RA is shown in Figure1, which shows the process of how RA stimulating consumption.

Figure1. The theoretical model

Reference

[1] Adomavicius, G., Tuzhilin, A.: Toward the Next Generation of Recommender Systems: A Survey of the State-of-the-art and Possible Extensions. IEEE Transactions on Knowledge and Data Engineering 17(6), 734–749 (2005).

[2] U. Gretzel, D.R. Fesenmaier, Persuasion in recommender systems, International Journal of Electronic Commerce 11 (2), 2007, pp. 81–100.

[3] A. Felfernig, B. Gula, An empirical study on consumer behavior in the interaction with knowledge-based recommender applications, in: Proceedings of the 8th IEEE International Conference on E-Commerce Technology and the 3rd IEEE International Conference on Enterprise Computing, E-Commerce, and E-Services (CEC/ EEE 06).

[4] Recommendation system http://en.wikipedia.org/wiki/Recommender_system

Social Network Analysis

1.     The definition and significance

Social network analysis (SNA) is an approach and set of techniques adopted to study and analyze the exchange of resources among individuals, teams, groups, departments, or even entire organizations and etc., which can be collectively named as actors (Haythornwaite, 1996). SNA provides a rich and systematic method to evaluate informal networks by mapping, measuring and analyzing relationships among actors (Cross, Parker, Prusak, & Borgatti, 2001).

SNA provides particular understandings about how actors connect and interact with each other in organization or in society, determines ways to consider diverse social structures, reveals issues and difficulties to measure phenomena. This relation analysis amongst numerous actors provides insight into communication links, locates the crucial actors in the information transmission and reveals structural holes in the social networks where the communication flows between nodes are inhibited. For instance, Rogers (2003) and Cross and Parker (2004) demonstrated the processes of building bridges amongst various subgroups in the organization, using SNA, to investigate how organizations became more efficient and sustainable (Valente 2010).
 

Figure1. Social Network Analysis
 

2.     Measurement

To assess the significance and contributions of each particular actor, SNA provides several measuring methods, in which, degree, closeness and centrality, as the most important and popular concepts, are detailed in the last lecture.

3.     An application of SNA – Pathological gambling

SNA is a powerful tool with increasing applications in social sciences and has been adopted in various areas (e.g., psychology, health, business organization and electronic communications) by examining the informal connections between factors.

Meisel, MK., Clifton, AD., MacKillop, J. etc. illustrated a research of how social network interacts with individuals on pathological gambling using SNA, in 2014. Will social factors initiate and maintain the behavior of gambling? The answer is yes, according to the paper. As shown in the two graphs (Figure2. (a)&(b)), darker colors and larger nodes stand for more frequent gambling, smoking and drinking behavior. Obviously, in PG network, a greater occurrence of gamblers is exhibited. On the contrary, the NPG participant comprised a network with few addictive behaviors, which are restricted to more distinct subgroups of associates
 

              

(a) Non-pathological gamblers (NPG) network’s gambling

 

(b) Pathological gamblers (PG) network’s gambling 

Figure2. Structural social networks of gambling

I personally believe that the social networks that we involved in deeply, would affects our thoughts, perspective, view, behavior, even our life.

 

References

[1] Haythornwaite, C., 1996, Social network analysis: An approach and technique for the study of information exchange, LISR, 18, 323–342.

[2] Cross, R., Parker, A., Prusak, L., & Borgatti, S. P., 2001, Knowing what we know: Supporting knowledge creation and sharing in social networks. Organizational Dynamics, 30, 100–120.

[3] Valente TE, 2010, Social Networks and Health, Models, Methods and Applications, Oxford University Press, Oxford.

[4] Meisel, MK., Clifton, AD., MacKillop, J. etc., 2014, Egocentric social network analysis of pathological gambling.

[5] Social network analysis [Online] http://wenku.baidu.com/link?url=5iakzIF2dQ1FiZiKSjUYxxZe-KyNg0dh8b5KtahIlZflqyAM5Ma-_op1R_kPhqbotbJ7A4cOTCDgSIRQSYwRSNhDI-yktOduw8KILKHRB3G Access at 16/Oct./2014.

[6] NodeXL [Online] http://research.microsoft.com/en-us/projects/nodexl/ Access at 16/Oct./2014.

Social Media Analytics -- Content Analysis

The information overload in social media becomes increasingly noticeable. This multitude of information raises the demand of content analysis, especially, information retrieval: extracting useful relevant information from a set of document. Lecture 3 demonstrated the main components of content analysis, which are document comparison, text classification and text clustering.

Document comparison

Document is constituted by numerous words in various orders, no doubt, it can be represented, analyzed and comparing by some particular words. The most representative word can be defined as the ones with high Term Frequency & Inverse Document Frequency (TF-IDF), the product of Term Frequency (number of times this word appears in this document) and Inverse Document Frequency
(). TF-IDF value increases proportionally to the frequency of a word in the document, but is offset by the occurrences of the term in the corpus, which attenuates the effect of terms that occur too often in almost every document.

Text classification

1.     Approaches

Two document classification approaches are listed as follows.

1)    Hierarchical categorization

Hierarchical categorization is a tree distribution, each document belongs to a unique category associated with it, and the categories are arranged hierarchically, as Figure 1. The most distinct drawback of this method could appear, when some elements not fit distinctly in one category.

Fugure1. Example of a hierarchical category

2)    Tag Categorization

Tag-based categorization is opposed to Hierarchical categorization, which means documents can be assigned to numerous related categorical tags.

2.     Technique

The text classification of new data can be implemented by a classifier, trained by using labeled data. Naïve Bayed text classification, the most popular sorting technique, classifies documents based on the frequency of representative words.

Text clustering

Text clustering can be briefly defined as grouping the similar objects in the data, by measuring their similarity using similarity functions. This clustering technique shows an outstanding contribution in the text domain, where the objects are of different granularities (such as documents, paragraphs, terms, etc.), especially in improving retrieval and supporting browsing.

Clustering can by generally divided, based on its results, into hard (one item to one cluster) and soft (one item to multiple clusters), K-Means Clustering and Model-based clustering are the most representative algorithms of these two clustering categories, respectively.

Reference：

[1] Klieber, P, Document classification through data mining social media networks, November, 2009.

[2] Aggarwal, CC, Zhao, Y, & Yu, PS, On text clustering with side information, ICDE Conference, 2012.