ABSTRACT
Contents 10.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 278 10.2 Recommender Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 279
10.2.1 Input Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 280 10.2.2 Recommender Algorithms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 281 10.2.3 Content-Based Filtering Algorithms. . . . . . . . . . . . . . . . . . . . . . . . . . . . 281
10.2.3.1 Naïve Bayes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 282 10.2.3.2 Latent Semantic Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . 283
10.2.4 Collaborative Algorithms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 284 10.2.4.1 Nonpersonalized Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . 286 10.2.4.2 Neighborhood-Based Algorithm . . . . . . . . . . . . . . . . . . . . 286 10.2.4.3 Latent Factors Collaborative Algorithms . . . . . . . . . . . . 287
10.2.5 Hybrid Algorithms. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 289 10.3 Evaluation of iTV Recommender Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 289
10.3.1 Quality Attributes. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 290 10.3.2 Subjective Tests . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 291 10.3.3 Off-Line Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 293 10.3.4 Online Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 294
10.4 Case Study . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 295 10.4.1 System Evaluation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 297
10.4.1.1 Off-Line Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 297 10.4.1.2 Online Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 300
10.5 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 302 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 304
10.1 Introduction Interactive television (iTV), differently from conventional television, allows providers (a) to track user activity and (b) to personalize the content transmitted to the users. These tasks are accomplished by recommender systems, whose goal is to filter information from a large dataset-e.g., thousands of channels and movies offered by an iTV operator-and to recommend to the users only the content that is likely of interest and attraction to them. Recommender systems play an important role in iTV services because the presence of a large number of TV programs dramatically reduces the visibility of each one, potentially inhibiting users from finding interesting TV contents. From the provider’s point of view, a large catalog is expensive to be maintained, due to the cost of the multimedia material itself, to the required storage space, and to the hardware infrastructure used to stream videos from the content provider to the users. Recently, many iTV recommender systems have been developed in academic as well as in corporate research labs, especially after the enormous resonance of the competition organized by the American movie rental provider Netflix. Differently from traditional e-commerce domains (e.g., iTunes, Last.fm, Amazon), recommender systems for iTV provide new challenges (e.g., realtime and scalability requirements, difficulties in collecting user ratings, difficulties in collecting TV content metadata).
