推薦系統會根據用戶行為、物品的評分反饋、或是其他資訊，分析以及預測用戶對物品的偏好或是評分。當推薦系統因為缺乏資訊而沒有辦法提出推薦時，我們就稱這個狀況為冷啟動問題。在推薦系統領域中，冷啟動問題一直都是一個挑戰。冷啟動問題有分成新物品、新用戶、新系統。新物品冷啟動問題指新物品進到推薦系統時，因為沒有反饋的紀錄提供給系統分析，所以無法被推薦出去。本論文針對新物品冷啟動問題，提出了一個稱為 ALFNCF (Attribute Latent Factor with Neural Collaborative Filtering)的混合推薦系統。混合推薦系統可以結合基於內容型推薦系統的優點，也就是善用內容資訊建立新舊物品的連結，同時也保留了協同過濾型推薦系統考慮到行為相似的用戶群或是獲得相似反饋的物品群之間互相影響的特點。ALFNCF結合物品屬性資訊，解決了因為系統僅考慮物品隱藏特徵，而新物品沒有評分記錄且難以學習的物品冷啟動問題。除了保留行為相似的用戶之間的互相影響外，還考慮了用戶、物品的屬性群體對個體的偏好影響。我們提出的方法除了保留了傳統隱藏特徵的線性交互影響和偏差值的考慮，還使用神經網路對隱藏特徵進行非線性分析，結合了兩者的優點。透過用戶過去對舊物品評分反饋的資訊來訓練模型，再用訓練好的模型準確預測出用戶對新物品的評分。經過實驗證明，我們提出的ALFNCF方法在預測新物品方面優於其他方法，也證明屬性資訊對於預測評分是有幫助的。

The recommendation system analyzes and predicts the user's preference or rating for items based on user behavior, rating feedback of items, or other information. When the recommender system cannot make recommendations due to lack of information, we call this situation a cold-start problem. The cold-start problem is a challenge for recommender systems. The cold-start problem is divided into new item, new user, and new system. The new item cold-start problem means that when a new item enters the recommender system, it cannot be recommended because there is no record of feedback. Aiming at the cold start problem of new items, this paper proposes a hybrid recommender system called ALFNCF (Attribute Latent Factor with Neural Collaborative Filtering). Hybrid recommendation system can combine the advantages of Content-based recommender system to make good use of content information to establish links between new and old items. It also retains the advantage of Collaborative filtering recommender system that considers the interaction between users with similar behaviors and items with similar feedback. ALFNCF combines item attribute information to solve the item cold-start problem that the system only considers the latent factors of the item, and the new item has no rating record and is difficult to learn. In addition to preserving the mutual influence between users with similar behaviors, the influence of users' and items' attribute groups on individual behaviors is also considered. Our method retains the linear interaction and deviation of a traditional hidden feature, and also nonlinearly analyzing the latent factors using the neural network, combining the advantages of both. Through the user's past rating feedback information training model for old items, the trained model is used to accurately predict the user's rating for new items. Experiments have proved that our proposed ALFNCF method is superior to other methods in predicting new items. It also proves that attribute information is helpful for predicting performance.

Verification Letter i
Acknowledgement ii
Chinese Abstract iii
Abstract iv
List of Figures vii
List of Tables viii
Chapter 1 Introduction 1
1.1. Research background and purpose 1
1.1.1. Recommender system 1
1.1.2. Cold-start problem 3
1.2. The structure of the paper 6
Chapter 2 Related Work 7
2.1. Traditional MF 7
2.2. Neural Network in recommendation 8
2.3. New item cold-start problem in recommendation 10
Chapter 3 Method 13
3.1. The input of attribute information 15
3.1.1. Binary vector input representation 16
3.1.2. Embedding layer 18
3.2. Linear part 19
3.2.1. The lack of item’s linear latent factor 20
3.2.2. Linear latent factor of attribute information 20
3.2.3. Bias consideration 21
3.2.4. Output of the linear part 22
3.3. Non-linear part 24
3.3.1. Non-linear latent factor of attribute information 25
3.3.2. Enhanced feature vector of item’s attribute 25
3.3.3. Output of the non-linear part 26
3.4. Whole architecture of ALFNCF 28
3.5. Example of ALFNCF 29
Chapter 4 Experiments 35
4.1. Experimental Setup 35
4.1.1. Datasets 35
4.1.2. Baselines 36
4.1.3. Evaluation criteria 38
4.2. Comparison with Baselines 39
4.3. Experiments for attribute information 44
4.3.1. Amount of item’s attribute information 44
4.3.2. Amount of user’s attribute information 47
4.4. Structure experiments 51
4.4.1. Linear part and non-linear part experiment 51
4.4.2. Linear latent factor of item conversion experiment 52
4.4.3. Convolution filter experiment 54
Chapter 5 Conclusion 56
References 57

[1] Breese, J. S., Heckerman, D., & Kadie, C. (2013). Empirical analysis of predictive algorithms for collaborative filtering. arXiv preprint arXiv:1301.7363.
[2] Desrosiers, C., & Karypis, G. (2011). A comprehensive survey of neighborhood-based recommendation methods. Recommender systems handbook, 107-144.
[3] Koren, Y., Bell, R., & Volinsky, C. (2009). Matrix factorization techniques for recommender systems. Computer, 42(8), 30-37.
[4] He, X., Liao, L., Zhang, H., Nie, L., Hu, X., & Chua, T. S. (2017, April). Neural collaborative filtering. In Proceedings of the 26th international conference on world wide web (pp. 173-182).
[5] Burke, R. (2002). Hybrid recommender systems: Survey and experiments. User modeling and user-adapted interaction, 12(4), 331-370.
[6] Rendle, S. (2010, December). Factorization machines. In 2010 IEEE International conference on data mining (pp. 995-1000). IEEE.
[7] Cheng, H. T., Koc, L., Harmsen, J., Shaked, T., Chandra, T., Aradhye, H., ... & Shah, H. (2016, September). Wide & deep learning for recommender systems. In Proceedings of the 1st workshop on deep learning for recommender systems (pp. 7-10).
[8] Park, S. T., & Chu, W. (2009, October). Pairwise preference regression for cold-start recommendation. In Proceedings of the third ACM conference on Recommender systems (pp. 21-28).
[9] Pazzani, M. J., & Billsus, D. (2007). Content-based recommendation systems. In The adaptive web (pp. 325-341). Springer, Berlin, Heidelberg.
[10] Guo, X., Yin, S. C., Zhang, Y. W., Li, W., & He, Q. (2019). Cold start recommendation based on attribute-fused singular value decomposition. IEEE Access, 7, 11349-11359.
[11] Enrich, M., Braunhofer, M., & Ricci, F. (2013, August). Cold-start management with cross-domain collaborative filtering and tags. In International Conference on Electronic Commerce and Web Technologies (pp. 101-112). Springer, Berlin, Heidelberg.
[12] Aleksandrova, M., Brun, A., Boyer, A., & Chertov, O. (2017). Identifying representative users in matrix factorization-based recommender systems: application to solving the content-less new item cold-start problem. Journal of Intelligent Information Systems, 48(2), 365-397.
[13] Funk, S. (2006). Netflix update: Try this at home.
[14] Koren, Y. (2008, August). Factorization meets the neighborhood: a multifaceted collaborative filtering model. In Proceedings of the 14th ACM SIGKDD international conference on Knowledge discovery and data mining (pp. 426-434).
[15] Koren, Y. (2009, June). Collaborative filtering with temporal dynamics. In Proceedings of the 15th ACM SIGKDD international conference on Knowledge discovery and data mining (pp. 447-456).
[16] Ma, H., Zhou, D., Liu, C., Lyu, M. R., & King, I. (2011, February). Recommender systems with social regularization. In Proceedings of the fourth ACM international conference on Web search and data mining (pp. 287-296).
[17] Guo, G., Zhang, J., & Yorke-Smith, N. (2015, February). Trustsvd: Collaborative filtering with both the explicit and implicit influence of user trust and of item ratings. In Proceedings of the AAAI Conference on Artificial Intelligence (Vol. 29, No. 1).
[18] Ji, K., & Shen, H. (2015). Addressing cold-start: Scalable recommendation with tags and keywords. Knowledge-Based Systems, 83, 42-50.
[19] Guo, H., Tang, R., Ye, Y., Li, Z., & He, X. (2017). DeepFM: a factorization-machine based neural network for CTR prediction. arXiv preprint arXiv:1703.04247.
[20] Covington, P., Adams, J., & Sargin, E. (2016, September). Deep neural networks for youtube recommendations. In Proceedings of the 10th ACM conference on recommender systems (pp. 191-198)..
[21] Zheng, L., Noroozi, V., & Yu, P. S. (2017, February). Joint deep modeling of users and items using reviews for recommendation. In Proceedings of the tenth ACM international conference on web search and data mining (pp. 425-434).
[22] Kim, D., Park, C., Oh, J., Lee, S., & Yu, H. (2016, September). Convolutional matrix factorization for document context-aware recommendation. In Proceedings of the 10th ACM conference on recommender systems (pp. 233-240).
[23] Dos Santos, C., & Gatti, M. (2014, August). Deep convolutional neural networks for sentiment analysis of short texts. In Proceedings of COLING 2014, the 25th International Conference on Computational Linguistics: Technical Papers (pp. 69-78).
[24] Ebesu, T., Shen, B., & Fang, Y. (2018, June). Collaborative memory network for recommendation systems. In The 41st international ACM SIGIR conference on research & development in information retrieval (pp. 515-524).
[25] Tang, J., & Wang, K. (2018, February). Personalized top-n sequential recommendation via convolutional sequence embedding. In Proceedings of the Eleventh ACM International Conference on Web Search and Data Mining (pp. 565-573).
[26] Melville, P., Mooney, R. J., & Nagarajan, R. (2002). Content-boosted collaborative filtering for improved recommendations. Aaai/iaai, 23, 187-192.
[27] Lam, X. N., Vu, T., Le, T. D., & Duong, A. D. (2008, January). Addressing cold-start problem in recommendation systems. In Proceedings of the 2nd international conference on Ubiquitous information management and communication (pp. 208-211).
[28] Volkovs, M., Yu, G. W., & Poutanen, T. (2017, December). DropoutNet: Addressing Cold Start in Recommender Systems. In NIPS (pp. 4957-4966).
[29] Pan, F., Li, S., Ao, X., Tang, P., & He, Q. (2019, July). Warm up cold-start advertisements: Improving ctr predictions via learning to learn id embeddings. In Proceedings of the 42nd International ACM SIGIR Conference on Research and Development in Information Retrieval (pp. 695-704).
[30] Li, J., Jing, M., Lu, K., Zhu, L., Yang, Y., & Huang, Z. (2019, July). From zero-shot learning to cold-start recommendation. In Proceedings of the AAAI Conference on Artificial Intelligence (Vol. 33, No. 01, pp. 4189-4196).
[31] Hu, L., Jian, S., Cao, L., Gu, Z., Chen, Q., & Amirbekyan, A. (2019, July). Hers: Modeling influential contexts with heterogeneous relations for sparse and cold-start recommendation. In Proceedings of the AAAI Conference on Artificial Intelligence (Vol. 33, No. 01, pp. 3830-3837).
[32] Qian, T., Liang, Y., Li, Q., & Xiong, H. (2020). Attribute graph neural networks for strict cold start recommendation. IEEE Transactions on Knowledge and Data Engineering.
[33] Glorot, X., Bordes, A., & Bengio, Y. (2011, June). Deep sparse rectifier neural networks. In Proceedings of the fourteenth international conference on artificial intelligence and statistics (pp. 315-323). JMLR Workshop and Conference Proceedings.
[34] Kingma, D. P., & Ba, J. (2014). Adam: A method for stochastic optimization. arXiv preprint arXiv:1412.6980.