Long short-term memory based movie recommendation
ABSTRACT
Recommender systems (RS) have become a fundamental tool for helping users make decisions
around millions of different choices nowadays – the era of Big Data. It brings a huge benefit for
many business models around the world due to their effectiveness on the target customers. A lot
of recommendation models and techniques have been proposed and many accomplished incredible outcomes. Collaborative filtering and content-based filtering methods are common, but these
both have some disadvantages. A critical one is that they only focus on a user's long-term static
preference while ignoring his or her short-term transactional patterns, which results in missing the
user's preference shift through the time. In this case, the user's intent at a certain time point may be
easily submerged by his or her historical decision behaviors, which leads to unreliable recommendations. To deal with this issue, a session of user interactions with the items can be considered as
a solution. In this study, Long Short-Term Memory (LSTM) networks will be analyzed to be applied
to user sessions in a recommender system. The MovieLens dataset is considered as a case study
of movie recommender systems. This dataset is preprocessed to extract user-movie sessions for
user behavior discovery and making movie recommendations to users. Several experiments have
been carried out to evaluate the LSTM-based movie recommender system. In the experiments,
the LSTM networks are compared with a similar deep learning method, which is Recurrent Neural
Networks (RNN), and a baseline machine learning method, which is the collaborative filtering using item-based nearest neighbors (item-KNN). It has been found that the LSTM networks are able
to be improved by optimizing their hyperparameters and outperform the other methods when
predicting the next movies interested by users
Trang 1
Trang 2
Trang 3
Trang 4
Trang 5
Trang 6
Trang 7
Trang 8
Trang 9
Tóm tắt nội dung tài liệu: Long short-term memory based movie recommendation
he training set. + Testing set is also used in evaluation. Specifically, unlike classification problem, evaluation in RS model plays the same role as unsupervised learning. In fact, there is no ensure that the list of movies recom- mended to user is correct or not. Thus, the used eval- uation method is a statistical measure to visualize the distribution of the testing test and validation test. Choosing a golden splitting ratio is an indispensable process in data splitting. It is usually called a cross val- idation process. The 80 – 20 or 90 – 10 (percentage) has been a golden ratio in both theoretical and prac- tical applications. In this study, the 80% for training data, 10% for testing data and 10% for validation data are applied for cross validation. Framework for detecting the appropriate data and evaluating the learning model in themovie recommender system In this study, the MovieLens dataset 13 with approxi- mately 10 million interactions is used. At the begin- ning, from the movie dataset, some features are cho- sen to build the proposed RS. Then, data preprocess- ing is applied to clean and convert user sessions to se- quences. After that, the dataset splitting is applied to define three sets. The training set is fed into themodel and do some tests to choose the best hyperparameter which can minimize the loss. Finally, the evaluation on the validation and testing sets by using Mean Re- ciprocal Rank (MRR) and other evaluation metrics as Precision@k, Recall@k and F1-Score@k is to confirm the used model is effective or not. The overall work- flow of detecting the best model and the evaluation is shown in Figure 3. In this study, Mean Reciprocal Rank (MRR) is used as a statistical method evaluating the model. In gen- eral, the Reciprocal Rank information retrieval mea- sure calculates the reciprocal of the rank at which the first relevant item was retrieved 14. When averaged across queries, the MRR is calculated. The formula of MRR is described as follows: MRR= 1 jQj jQj å i=1 1 ranki Where ranki refers to the rank position of the first rel- evant item in the ith query; Q is the number of items. The model computes MRR score for both validation and testing. Then, the model is evaluated to be good when MRR scores given on both testing and vali- dation set is approximately same. Other evaluation metrics used in this approach is Precision@k and Re- call@k14. In comparison to MRR, these metrics care on k highest ranking items, which are the reasonable evaluation measures for emphasizing returning more relevant items earlier. The key point of this method is SI4 Science & Technology Development Journal – Engineering and Technology, 3(S1):SI1-SI9 to take all the precisions at k for each the sequence in the testing set. More details, the sequence of length n is splitted into two parts: the sequence of length k for comparison and the other sequence of length n – k put into the predicting function, and then the sorted fac- torization matrix scores are retrieved. If any item in top k highest scores matches the one in the sequence of length k, the number of hits is increased by 1. Then, the precision at k for one sequence of length n is given by the number of hits divided by k, which stands for the number of recommended items. For the recall, k stands for the number of relevant items. In facts, k in recall is usually smaller than the one in precision. Fi- nally, the mean average precision and recall at k are calculated for all sequences in the testing set. In gen- eral, the formulas of the precision, recall and F1-score at k are described as follows. Precision@k = relevant_items in top kk Recall@k = relevant_items in top krelevant_items F1 score@k = 2 Precision@kRecall@kPrecision@k+Recall@k Figure 3: Overall workflow of detecting the best model for the movie recommender system LSTMmodel Optimization Hyperparameter optimization for LSTM model. LSTM hyperparameter There is not a good methodology to choose an ideal hyperparameter for the neural network up to now. Thus, the more trials, the better results are for the model. In this study, an automatically testing pro- gramwith some randomhyperparameters is built and takes three days consecutively to find the best hy- perparameter. Some hyperparameters used in the configuration are embedding dimension, number of epochs, random state (shuffling number of interac- tions), learning rate, batch size... The loss function is kept same in the experiments. Loss function Several loss functions are applied to find the most ap- propriatemodel, the formulas of them are listed in Ta- ble 1. Model efficiency evaluation In this study, LSTM, RNN and another baseline method are chosen to compare the evaluation met- rics. The common baseline is Item-KNN, which con- siders the similarity between the vectors of sessions. This baselinemethod is one of themost popular item- to-item solutions in practical systems. The MRR, Av- erage Precision, Average Recall at 20 are measured to find out the efficiency of the LSTM versus the others. EXPERIMENTAL RESULTS AND EVALUATION Hyperparameter optimization The experiment is taken by running 10 trials on the randomly selected hyperparameters which are de- fined in the fixed list as follows: + Learning rate: [0.001, 0.01, 0.1, 0.05] + l2: [1e-6, 1e-5, 0, 0.0001, 0.001] + Embedding dimension: [8, 32, 64, 128, 256] + Batch size: [8, 16, 32, 64, 128] The loss function used in this experiment is BPR and the number of epochs for each trail equals 10. The loss sorted results is shown in Table 2. The best result of the experiment is chosen for the model. In facts, the model requires a big system for training faster with more epochs, but the current one is not enough for training longer. Therefore, there should be more time for training a complete model. Loss function This experiment compares the efficiency of some loss functions mentioned in the previous section. The hy- perparameter is chosen from the best one in the pre- vious experiment. The results of the experiment on the training set in four types of loss after 10 epochs are illustrated in Figure 4. SI5 Science & Technology Development Journal – Engineering and Technology, 3(S1):SI1-SI9 Table 1: Loss function formulas for themodel Pointwise L= p positive_loss+negative_loss positive_loss = 1 - sigmoid(pos_pred) negative_loss=sigmoid(neg_pred) Hinge L= p maxf0; (neg_pred pos_pred) +1g Adaptive Hinge L= p maxf0; (neg_pred hightest_pos_pred) +1g Bayesian Personalized Ranking (BPR) L= p 1 sigmoid(pos_pred neg_pred) neg_pred: negative_prediction; pos_pred: positive_prediction Figure 4: Loss function performance on the model Table 2: Sorted loss results for several hyperparameters test Batch size Embedding dimen- sion Learning rate L2 loss 128 64 0.001 0.0001 0.1962 64 64 0.01 0.0001 0.2491 8 8 0.01 1e-6 0.2537 16 256 0.1 1e-5 0.2701 16 32 0.05 0.001 0.271 16 128 0.01 0 0.276 16 64 0.001 1e-6 0.281 32 32 0.05 1e-5 0.2944 16 128 0.1 0.001 0.3566 8 128 0.05 0 0.4204 According to the graph in Figure 4, theBPRandHinge loss can minimize the loss better than others can. Es- pecially BPR, which can run the loss well as the be- ginning and it looks more stable than Hinge in reg- ularization, as the comparison in training and test- ing in Figure 5. Therefore, BPR is chosen to perform the model instead of Hinge, Pointwise and Adaptive Hinge. Evaluation results The evaluation results between LSTM, RNN and the baseline Item-KNNmethod is shown in Table 3. According to the experiment results, both LSTM and RNN can perform better than the baseline method (Item-KNN).Moreover, LTSM performs well in building a RS with specific relevant movies. LSTM can forget what it thinks to be not necessary for the long-term. Therefore, the learning results of LSTM are more updatable by time and frequency of interac- tions. Overall, LSTM is a better model for building the session-based RS. SI6 Science & Technology Development Journal – Engineering and Technology, 3(S1):SI1-SI9 Figure 5: Good fit on training and testing by BPR loss function Table 3: RNN on the testing set DLModel Precision @20 Recall @20 F1-Score@20 MRR (test/validation) LSTM 0.705 0.707 0.706 0.239/0.235 RNN (dropout =0.2) 0.598 0.614 0.606 0.224/0.204 Item-KNN 0.506 0.507 0.506 0.201/0.206 DISCUSSION In our approach, the modernmodel of recurrent neu- ral networks, i.e., LSTM, is applied to do themovie RS with the task of session-based recommendations. Be- sides, themodification of LSTM in order to fit it better is performed by using session-parallel mini-batches and ranking losses. The evaluation results have shown the outstanding improvement in comparison with the popular baseline approach. The movie dataset pro- vided by GroupLens is excellent for researching on some principal features. Thus, this approach can be applied not only for movie data, but also for some other practical fields. CONCLUSIONS In conclusions, the LSTM-based movie RS has been proposed and can achieve higher recommendation performance when optimizing the hyperparameters of LSTM, using loss function and optimization func- tion. The loss is calculated to keep decreasing af- ter each epoch and keep as minimum as possible for the long-term computation. Adam optimizer plays a great role in modifying the hyperparameter. More- over, LSTM has been proved as the better model than RNN during evaluation. Despite the results of both are not good enough, but this study has presented some solutions to improve the accuracy of the learn- ing model. ACKNOWLEDGMENTS This research is funded by Vietnam National Foun- dation for Science and Technology Development (NAFOSTED) under grant number: 06/2018/TN. ABBREVIATIONS RS: Recommender Systems LSTM: Long Short-Term Memory RNN: Recurrent Neural Networks DL: Deep Learning KNN: K-Nearest Neighbors BPR-MF: Bayesian Personalized Ranking – Matrix Factorization SI7 Science & Technology Development Journal – Engineering and Technology, 3(S1):SI1-SI9 SGD: Stochastic Gradient Descent MRR: Mean Reciprocal Rank COMPETING INTERESTS The authors hereby declare that there is no conflict of interest in the publication of the article. AUTHORS’ CONTRIBUTION Duy Bao Tran is involved in proposing and imple- menting solutions, and writing reports. Thi Thanh Sang Nguyen has giving ideas and so- lutions, assess experimental results and writing the manuscript. REFERENCES 1. Aggarwal CC. Recommender Systems. Springer International Publishing. 2016;Available from: https://doi.org/10.1007/978- 3-319-29659-3. 2. Agrawal R, Srikant R. Fast Algorithms for Mining Association Rules in Large Databases. Proceedings of the 20th Interna- tional Conference on Very Large Data Bases. 1994;p. 487–499. 3. Nguyen TTS, Lu H, Tran TP, Lu J. Investigation of Sequen- tial Pattern Mining Techniques for Web Recommendation. International Journal of Information and Decision Sciences (IJIDS). 2012;p. 293–312. Available from: https://doi.org/10. 1504/IJIDS.2012.050378. 4. Vieira A. Predicting online user behaviour using deep learning algorithms. arXiv:151106247v3. 2016;. 5. Devooght R, Bersini H. Collaborative Filtering with Recurrent Neural Networks. arXiv:160807400. 2016;. 6. Hidasi B, Karatzoglou A, Baltrunas L, Tikk D. Session-based Recommendations with Recurrent Neural Networks. 2015;. 7. GravesA. Supervised Sequence LabellingwithRecurrentNeu- ral Networks. Springer-Verlag Berlin Heidelberg. 2012;Avail- able from: https://doi.org/10.1007/978-3-642-24797-2. 8. Patterson J, Gibson A. Deep Learning-A Practitioner’s Ap- proach. O’Reilly Media, Inc. 2017;. 9. Rendle S, Freudenthaler C, Gantner Z, Schmidt-Thieme L. BPR: Bayesian personalized ranking from implicit feedback. Pro- ceedings of the Twenty-Fifth ConferenceonUncertainty inAr- tificial Intelligence, Montreal, Quebec, Canada,. 2009;p. 452– 461. 10. KingmaDP, Ba J. Adam: AMethod for StochasticOptimization. 2015;. 11. Duchi J, Hazan E, Singer Y. Adaptive Subgradient Methods for Online Learning and Stochastic Optimization. J Mach Learn Res. 2011;12:2121–2159. 12. Hinton G, Srivastava N, Swersky K. Lecture 6d- a separate, adaptive learning rate for each connection. Slides of Lecture Neural Networks for Machine Learning. 2012;. 13. Grouplens. 2019;Available from: https://grouplens.org/ datasets/movielens/. 14. Liu L, Ozsu MT. Encyclopedia of Database Systems. Springer. 2009;Available from: https://doi.org/10.1007/978-0- 387-39940-9. SI8 Tạp chí Phát triển Khoa học và Công nghệ – Engineering and Technology, 3(S1):SI1-SI9 Open Access Full Text Article Bài Nghiên cứu Khoa Công nghệ Thông tin, Trường Đại học Quốc tế, ĐHQG-HCM, Việt Nam Liên hệ Nguyễn Thị Thanh Sang, Khoa Công nghệ Thông tin, Trường Đại học Quốc tế, ĐHQG-HCM, Việt Nam Email: nttsang@hcmiu.edu.vn Lịch sử Ngày nhận: 10-8-2019 Ngày chấp nhận: 22-8-2019 Ngày đăng: 19-9-2019 Bản quyền © ĐHQG Tp.HCM. Đây là bài báo công bố mở được phát hành theo các điều khoản của the Creative Commons Attribution 4.0 International license. Đề xuất phim dựa trên bộ nhớ ngắn hạn Trần Duy Bảo, Nguyễn Thị Thanh Sang* Use your smartphone to scan this QR code and download this article TÓM TẮT Hiện nay, các hệ thống đề xuất đã trở thànhmột công cụ cơ bản để giúp người dùng đưa ra quyết định trong hàng triệu lựa chọn khác nhau - kỷ nguyên của Dữ liệu lớn. Nó mang lại lợi ích rất lớn cho nhiều mô hình kinh doanh trên toàn thế giới do hiệu quả của chúng đối với khách hàng. Rất nhiều mô hình và kỹ thuật khuyến nghị đã được đề xuất và có nhiều kết quả đáng kinh ngạc. Phương pháp lọc cộng tác và phương pháp lọc dựa trên nội dung là phổ biến, nhưng cả hai đều có một số nhược điểm. Một điều quan trọng là chúng chỉ tập trung vào sở thích tĩnh dài hạn của một người dùng trong khi bỏ qua các mẫu giao dịch ngắn hạn, dẫn đến bỏ sót sự thay đổi sở thích của người dùng trong suốt thời gian. Trong trường hợp này, mối quan tâm của người dùng ở một thời điểm nhất định có thể dễ dàng che mờ bởi các hành vi quyết định trong lịch sử của người đó, dẫn đến các khuyến nghị không đáng tin cậy. Để giải quyết vấn đề này, một phiên tương tác của người dùng với các mục có thể được coi là một giải pháp. Trong nghiên cứu này, các mạng Bộ nhớ ngắn hạn (LSTM) sẽ được phân tích để áp dụng cho các phiên của người dùng trong hệ thống đề xuất. Bộ dữ liệu MovieLens được dùng trong nghiên cứu hệ thống tư vấn phim. Một số thí nghiệm được thực hiện để đánh giá hệ thống đề xuất phim dựa trên LSTM. Từ khoá: Học sâu, Bộ nhớ ngắn hạn, Hệ thống đề xuất, Khai thác chuỗi Trích dẫn bài báo này: Bảo T D, Sang N T T. Đề xuất phim dựa trên bộ nhớ ngắn hạn. Sci. Tech. Dev. J. -Eng. Tech.; 3(S1):SI1-SI9. SI9 DOI : 10.32508/stdjet.v3iSI1.540
File đính kèm:
- long_short_term_memory_based_movie_recommendation.pdf