=Paper=
{{Paper
|id=Vol-1670/paper-22
|storemode=property
|title=None
|pdfUrl=https://ceur-ws.org/Vol-1670/paper-22.pdf
|volume=Vol-1670
}}
==None==
<pdf width="1500px">https://ceur-ws.org/Vol-1670/paper-22.pdf</pdf>
<pre>
  Applying Topic Model in Context-Aware TV
          Programs Recommendation

                   Jing Yuan1 , Andreas Lommatzsch1 , Mu Mu2
               1
               DAI-Labor, Technische Universität Berlin, Germany
                 {jing.yuan, andreas.lommatzsch}@dai-labor.de
       2
         The University of Northampton, UK – mu.mu@northampton.ac.uk


      Abstract. In IPTV systems, users’ watching behavior is influenced by
      contextual factors like time of day, day of week, Live/VOD condition
      etc., yet how to incorporate such factors into recommender depends on
      the choice of basic recommending model. In this paper, we apply a topic
      model in Information Retrieval (IR)–Latent Dirichlet Allocation (LDA)
      as the basic model in TV program recommender. What makes employ-
      ing such approach meaningful is the resemblance between user watching
      frequency as the entry in user-program matrix and term frequency in
      term-document matrix. In addition, we propose an extension to this user-
      oriented LDA by adding a probabilistic selection node in this probabilis-
      tic graphical model to learn contextual influence and user’s individual
      inclination on di↵erent contextual factors.
      The experiment using the proposed approach is conducted on the data
      from a web-based TV content delivery system “Vision”, which serves
      the campus users in Lancaster University. The experimental results show
      that both user-oriented LDA and context-aware LDA converge smoothly
      on perplexity regarding both iteration epoch and topic numbers under in-
      ference framework Gibbs Sampling. In addition, context-aware LDA can
      perform better than user-based LDA and baseline approach on both pre-
      cision metrics and diversity metrics when the number of topic is over 50.
      Aside from that, programs with highest probability distribution within
      top 10 topics represent the natural clustering e↵ect of applying this topic
      model in TV recommender.


Keywords: TV recommender, context-awarenes, Latent Dirichlet Allocation.

Resubmission of J. Yuan, F. Sivrikaya, F. Hopfgartner, A. Lommatzsch, and
M. Mu. Context-aware LDA: Balancing relevance and diversity in TV content
recommenders. In Proc. of the 2nd Workshop on RecsysTV, Sept. 2015.

Acknowledgement The work of the first author has been continuously funded
by China Scholarship Council (CSC).

  Copyright c 2016 by the paper’s authors. Copying permitted only for private and aca-
  demic purposes. In R. Krestel, E Müller: Proceedings of the LDWA 2016, Workshop:
  KDML, Potsdam, Germany, 12-14 September 2016, published at http://ceur-ws.org

</pre>