Recommendation of CGM novels considering serendipity.

Size: px

Start display at page:

Download "Recommendation of CGM novels considering serendipity."

Fay Chapman
5 years ago
Views:

1 CGM Recommendation of CGM novels considering serendipity Kyushu University Abstract: Recently, CGM (Consumer Generated Media) services become popular. Although a huge amount of contents have been posted to CGM site, but only a few contents are selected and viewed by users. This phenomenon are called as very short head and very long tails. Some researcher mentioned that recommendation algorithms made influence to user s content selection strongly. Serendipitous recommendation is necessary for healthy cultural glowth of CGM contents. In this paper, we propose a new serendipitious recommendation method for CGM novels. We apply clustering to novels to divide into middle size clusters, and contents are similar in the same cluster. For clustering, each content is represented into a vector using doc2vec. Next, we calculate the distance from the user s preference (bookmark) to each clusters. We believe that contents in near but not nearest cluster are better for serendipitious recommendation. We apply our method to web CGM novels in syosetu.com, and we also construct a web based recommendation system. 1 CGM, Consumer Generated Media) CGM Youtube pixiv CGM CGM Web CGM t.iida.630@s.kyushu-u.ac.jp

2 2 20 [3] 20 Amazon [1] Amazon (user) (item) SVD (Singular Value Decomposition) Matrix Factorization Steffen Rendle Factorization Machines (FM) [4] FM Mouzhi [5] Himan [2] 80% 80% 2 Himan 3 Web CGM 3.1 ( Wikipedia [6] PV ,185, , API Web API API Web API Python ,260 JSON YAML, 1 JSON SQLite3 DB API HTML Web Python URL ID ID ncode 10

3 1: ncode ID title writerid ID writer story big genre genre keyword general firstup general lastup novel type isend general point fav novel cnt review cnt all point all hyoka cnt sasie cnt taken data ,123 29,768,817 SQLite3 DB 4 web web 5 CGM CGM 2 A A [7] 2 web API Doc2Vec Ward Ward

4 5.1 Doc2Vec Word2Vec Doc2Vec Word2Vec Tomas Mikolov [8] Word2Vec Countinuous Bag-of-Words Skip-gram Hierarchical Softmax Negative Sampling Doc2Vec [9] Doc2Vec 5.2 Ward N N Ward A B D(A, B) D(A, B) = d(x, µ AB ) 2 d(x, µ A ) 2 + d(x, µ B ) 2 x A x B x A,B = S AB (S A + S B ) (1) d(x, y) µ AB A B µ A µ B A B S Web CGI 72 E c E c s 1,..., s i,..., s k k C i C i m i E c N k i=1 E c = (s k i m i ) i=1 k i=1 s = (s i m i ) (2) i N Doc2Vec Word2Vec Doc2Vec Doc2Vec [10] Python gensim [11] Doc2Vec (100 ) Python Scipy [12] Ward MeCab

5 E c E c E c userid clustering C = C #,, C & clustering_result bookmark bookmarkdb send user s bookmark B = n #,, n & distance d(b, C & ) novel noveldb 1: C: クラスタ集合 C & : クラスタ i に含まれる説集合 B: ブックマーク集合 n &: 説 d(b, C & ): B と C & の距離 Recommendation k ncode csv ID ID ncode ncode B C 0, C 1,..., C k B B C k C k B CGM CGM web Ward 0 [1] Brent Smith and Greg Linden. Two decades of recommender systems at amazon.com. IEEE Internet Computing, Vol. 21, No. 3, pp , May-June 2017.

6 [2] Himan Abdollahpouri, Robin Burke, and Bamshad Mobasher. Controlling popularity bias in learning to rank recommendation. In ACM RecSys17 (Proceedings of the Eleventh ACM Conference on Recommender Systems), RecSys 17, pp , August [3] Paul Resnick and Hal R. Varian. Recommender systems. Communications of the ACM, Vol. 40, No. 3, pp , March [4] Steffen Rendle. Factorization machines. In ICDM 10 (Proceedings of the 2010 IEEE International Conference on Data Mining), pp , December [5] Mouzhi Ge, Carla Delgado-Battenfeld, and Dietmar Jannach. Beyond accuracy: Evaluating recommender systems by coverage and serendipity. In ACM RecSys 10 (Proceedings of the fourth ACM conference on Recommender systems), pp , September [6] Wikipedia. in wikipedia. E8%AA%AC%E5%AE%B6%E3%81%AB%E3%81%AA%E3% 82%8D%E3%81%86. [7]. niconico. watch/kn3440, [8] Tomas Mikolov, Ilya Sutskever, Kai Chen, Greg Corrado, and Jeffrey Dean. Distributed representations of words and phrases and their compositionality. In Proceedings of the 26th International Conference on Neural Information Processing Systems, Vol. 2 of NIPS 13, pp , USA, Curran Associates Inc. [9] Quoc Le and Tomas Mikolov. Distributed representations of sentences and documents. In Proceedings of the 31st International Conference on Machine, pp , [10],. Cgm. 29, pp , [11] gensim topic modeling for humans. radimrehurek.com/gensim/. [12] E. Jones, T. Oliphant, P. Peterson, and et al. Scipy: Open source scientific tools for python.

Man is to Computer Programmer as Woman is to Homemaker? Debiasing Word Embeddings. Md Musa Leibniz University of Hannover

Man is to Computer Programmer as Woman is to Homemaker? Debiasing Word Embeddings Md Musa Leibniz University of Hannover Agenda Introduction Background Word2Vec algorithm Bias in the data generated from