Hybrid Recommendations

Table 4.7: Performance on complete feature vector

Features MAP@10 Recall Note

U + S 0.3817 0.5216 Base-line

U + S + C* 0.5120 0.6614

U + S + C* + S* 0.5236 0.6684 Hybrid

U + S + C* + S* 0.5251 0.6708 Hybrid + Grouping Note: C* denotes all the categorical features, and S* de-notes all the extracted similarity features.

Table 4.8: Coverage scores with top-k similarity information

Features Coverage

We studied if we can further boost the accuracy by integrating all the proposed similarity features, including categorical ones (denoted as C* collectively) and similarity features (denoted as S* collectively). As Table 4.7 shows, using more data generally leads to bet-ter accuracy. When all the features are considered (U+S+C*+S*), we are able to obtain 0.5251 in MAP@10 and 0.6708 in recall, both of which are the highest ones in our eval-uation. This result confirms again the ability of the proposed method in incorporating multiple similarity information.

4.5 Recommendation Diversity

Every user has his/her own habit of music listening. In other words, the motivation of choosing a type of music to listen to may depends on user’s mood, the weather or the oc-casion. That is why we generally consider the utilized features as more as possible while building the recommender system. From a different perspective to handle the problem, some users may want to discover the musics they never know before. Base on such sit-uation, we seek to investigate the recommendation diversity in this section. The goal is to recommend the diverse songs to the user who only focus on specific domain of music.

•‧

Figure 4.5: Recommendation diversity with top k similarity information

In our experiments, the basis of diversity rests on a k-means clustering result of audio information. Consequently, all the musics are partitioned into 10 clusters according to the 53-dimensional audio feature. The clustering results are denoted as K in our exper-iments. Finally, we filter out the users who already listened to various clusters, because our objective is to help the users discovering the types of song they never listen to before.

Similarity and Diversity are two sides of one thing. By integrating different numbers of similarity information, our proposed method is able to control how diverse the rec-ommended musics are. Table 4.8 lists the coverage scores while different numbers of similarity information had been considered. The result show that we build the recom-mendation results with coverage score from 0.19 to 0.55. Moreover, the lower amount of similarity information we use, the lower coverage score we get. The higher amount of similarity information we user, the higher coverage score we get. As the growth of similarity information, the proposed method is able to find out more domains of music.

Figure 4.5 can help us finding the tendency via comparing to the base-line result (i.e., U + S + C). It implies that controlling the number of similarity information can indirectly control the recommendation diversity.

28

•‧

國

立立 政 治 大

㈻㊫學

•‧

N a tio na

l C h engchi U ni ve rs it y

Chapter 5 Conclusions

In this paper, we have presented a novel approach that incorporates multiple feature simi-larity to factorization model via feature engineering. The simisimi-larity computation captures the similar patterns from the objects and enhances the convergence speed and accuracy of FM. The proposed method is applicable to many kinds of features, which means we can obtain the higher level information from multiple aspects. Our experimental results show that feature similarity indeed benefits the recommendation performance. We also pro-pose several features, including CF-based, content-based and context-based ones. Among these features, we try to capture the relationship between users and songs by matching users’ emotions. The results show that the idea is able to enhance the quality of recom-mendations. Moreover, by integrating different number of similarity information, we are capable of controlling how diverse the recommendations are. Then, in order to avoid the noise within large similarity features, we adopt the grouping FM as the extended method to model the problem. The unnecessary connection can be eliminated if the features are within a same group. With the aforementioned technical contributions, we are able to im-prove the Mean Average Precision in music recommendation for a real-world, large-scale dataset from 0.3817 to 0.5251, comparing to the tradition CF-based baseline. In addi-tion, via combing different number of similarity informaaddi-tion, we can make the prediction results from 0.19 to 0.55 in terms of Coverage.

Diversity becomes an important issue for real-world Recommendation Systems. Al-though our proposed similarity features can diversify the prediction results, to go one step further, we seek to start the work from the core of model. That is, we are looking for-ward about developing a recommendation model that is capable of generating a diverse recommendation list without feature engineering.

•‧

國

立立 政 治 大

㈻㊫學

•‧

N a tio na

l C h engchi U ni ve rs it y

30

•‧

[1] Factorization machines with libfm. ACM Trans. Intell. Syst. Technol., 3(3):57:1–

57:22, May 2012.

[2] R. Agrawal, S. Gollapudi, A. Halverson, and S. Ieong. Diversifying search results.

In Proceedings of the Second ACM International Conference on Web Search and Data Mining, WSDM ’09, pages 5–14. ACM, 2009.

[3] F. Aiolli. A preliminary study of a recommender system for the million songs dataset challenge. In Proceedings of the ECAI Workshop on Preference Learning: Problems and Application in AI, 2012.

[4] M. Bradley and P. J. Lang. Affective norms for english words ANEW: Instruction manual and affective ratings. Technical report, The Center for Research in Psy-chophysiology, Univ. Florida, 1999.

[5] R. Cai, C. Zhang, C. Wang, L. Zhang, and W.-Y. Ma. Musicsense: contextual music recommendation using emotional allocation modeling. In Proceedings of the 15th international conference on Multimedia, MULTIMEDIA ’07, pages 553–556.

ACM, 2007.

[6] K. Chen, T. Chen, G. Zheng, O. Jin, E. Yao, and Y. Yu. Collaborative personal-ized tweet recommendation. In Proceedings of the 35th international ACM SIGIR conference on Research and development in information retrieval, SIGIR ’12, pages 661–670. ACM, 2012.

[7] T. Chen, L. Tang, Q. Liu, D. Yang, S. Xie, X. Cao, C. Wu, E. Yao, Z. Liu, Z. Jiang, et al. Combining factorization model and additive forest for collaborative followee recommendation.

[8] J. Davidson, B. Liebald, J. Liu, P. Nandy, T. Van Vleet, U. Gargi, S. Gupta, Y. He, M. Lambert, B. Livingston, and D. Sampath. The youtube video recommendation system. In Proceedings of the fourth ACM conference on Recommender systems, RecSys ’10, pages 293–296. ACM, 2010.

•‧

[9] D. T. Derek Tingle, Youngmoo E. Kim. Exploring automatic music annotation with acoustically-objective tags. pages 55–61, 2010.

[10] N. Hariri, B. Mobasher, and R. Burke. Context-aware music recommendation based on latenttopic sequential patterns. In Proceedings of the sixth ACM conference on Recommender systems, RecSys ’12, pages 131–138. ACM, 2012.

[11] L. Hong, A. S. Doumith, and B. D. Davison. Co-factorization machines: modeling user interests and predicting individual decisions in twitter. In Proceedings of the sixth ACM international conference on Web search and data mining, WSDM ’13, pages 557–566. ACM, 2013.

[12] M. Jiang, P. Cui, R. Liu, Q. Yang, F. Wang, W. Zhu, and S. Yang. Social contex-tual recommendation. In Proceedings of the 21st ACM international conference on Information and knowledge management, CIKM ’12, pages 45–54. ACM, 2012.

[13] M. Jiang, P. Cui, F. Wang, Q. Yang, W. Zhu, and S. Yang. Social recommendation across multiple relational domains. In Proceedings of the 21st ACM international conference on Information and knowledge management, CIKM ’12, pages 1422–

1431. ACM, 2012.

[14] N. Koenigstein, G. Dror, and Y. Koren. Yahoo! music recommendations: modeling music ratings with temporal dynamics and item taxonomy. In Proceedings of the fifth ACM conference on Recommender systems, RecSys ’11, pages 165–172. ACM, 2011.

[15] Y. Koren. Collaborative filtering with temporal dynamics. In Proceedings of the 15th ACM SIGKDD international conference on Knowledge discovery and data mining, KDD ’09, pages 447–456. ACM, 2009.

[16] G. Linden, B. Smith, and J. York. Amazon.com recommendations: Item-to-item collaborative filtering. IEEE Internet Computing, 7(1):76–80, Jan. 2003.

[17] H. Ma, C. Liu, I. King, and M. R. Lyu. Probabilistic factor models for web site recommendation. In Proceedings of the 34th international ACM SIGIR conference on Research and development in Information Retrieval, SIGIR ’11, pages 265–274.

ACM, 2011.

[18] I. Pil´aszy and D. Tikk. Recommending new movies: even a few ratings are more valuable than metadata. In Proceedings of the third ACM conference on Recom-mender systems, RecSys ’09, pages 93–100. ACM, 2009.

32

•‧

國

立立 政 治 大

㈻㊫學

•‧

N a tio na

l C h engchi U ni ve rs it y

[19] S. Rendle. Bayesian factorization machines.

[20] S. Rendle. Factorization machines. In Proceedings of the 10th IEEE International Conference on Data Mining. IEEE Computer Society, 2010.

[21] S. Rendle, Z. Gantner, C. Freudenthaler, and L. Schmidt-Thieme. Fast context-aware recommendations with factorization machines. In Proceedings of the 34th international ACM SIGIR conference on Research and development in Information Retrieval, SIGIR ’11, pages 635–644. ACM, 2011.

[22] T. Sakai. Evaluation with informational and navigational intents. In Proceedings of the 21st international conference on World Wide Web, WWW ’12, pages 499–508.

ACM, 2012.

[23] T. Sakai and R. Song. Evaluating diversified search results using per-intent graded relevance. In Proceedings of the 34th international ACM SIGIR conference on Re-search and development in Information Retrieval, SIGIR ’11, pages 1043–1052.

ACM, 2011.

[24] Y. Shen and R. Jin. Learning personal + social latent factor model for social recom-mendation. In Proceedings of the 18th ACM SIGKDD international conference on Knowledge discovery and data mining, KDD ’12, pages 1303–1311. ACM, 2012.

[25] Y. Shi, A. Karatzoglou, L. Baltrunas, M. Larson, A. Hanjalic, and N. Oliver. Tfmap:

optimizing map for top-n context-aware recommendation. In Proceedings of the 35th international ACM SIGIR conference on Research and development in infor-mation retrieval, SIGIR ’12, pages 155–164. ACM, 2012.

[26] J. Weston, C. Wang, R. Weiss, and A. Berenzweig. Latent collaborative retrieval.

CoRR, abs/1206.4603, 2012.