結合意見分析與使用者喜好之評論分數預測

Rating prediction based on combination of opinion analysis and user preference from textual reviews

許家瑜(Chia-Yu Hsu)

指導教授 ： 賴錦慧

中原大學/商學院/資訊管理研究所/碩士(2018年)

https://doi.org/10.6840/cycu201800471

摘要

現在有許多兼具社群功能的評論網站，讓使用者可以分享使用評論與心得、給予商家評價等，也使得網路資料大量成長，更讓使用者面臨資訊過載的問題。許多研究提出推薦方法來解決資訊過載的問題。然而，社群評論資料為非結構化文字資料，且包含不同面向的潛在資訊，而各面向對不同使用者與商家有不同重要性。若是在評分預測時只分析使用者的數值評分資料，無法得知使用者相關之面向與各面向的重要性，亦無法得知商家在各面向之評價，可能造成預測失準的狀況。
為了解決上述問題，本研究提出一個以面向分析為基礎之評分預測方法，整合使用者在不同面向的喜好、商家在各面向之表現、以及使用者在網站中潛在社群關係，再利用社群矩陣分解 (Social Matrix Factorization)進行評分預測，以預測使用者未來可能感興趣、且符合使用者面向特徵之商家。
經實驗結果可知，本研究所提出的以面向分析為基礎之評分預測方法，相較於其他傳統以評分為基礎的評分預測方法，可以有效的利用評論文字針對不同面向進行分析，並提高評分預測的準確度。

關鍵字

資料探勘 ； 評論探勘 ； 自然語言處理 ； 面向偵測 ； 評分預測 ； 矩陣分解

並列摘要

The user review websites allows users to share their reviews of products or businesses, give ratings to products or businesses, and interact with other users. Because the rapid growth of online review data, users face the information overload problem. To resolve such problem, many researches proposed various recommendation methods based on the analysis of users’ ratings. Besides users’ ratings, the reviews of products or businesses is unstructured textual data and contain the information of different aspects. These aspects have different impact and importance to both users and businesses. It may lead inaccurate rating predictions because it is difficult to know users’ related aspects and their corresponding importance by analyzing users’ rating data.
To resolve the above problems, this research proposes a novel rating prediction method based on aspect analysis, which integrates users’ preferences and businesses’ performance in different aspects, and users’ social relationships in the website. Then, the social matrix factorization method is used to predict the ratings for users on businesses, which users may be interested in and have similar aspects features with users’ aspects.
Based on the experimental results, the proposed method performs better than other traditional rating-based prediction methods. Our method can effectively analyze various aspects from the reviews of users and businesses and can improve the accuracy of rating predictions.

並列關鍵字

Data Mining ； Opinion Mining ； Natural Language Processing ； Aspect Detection ； Rating Prediction ； Matrix Factorization

參考文獻

[1] A. Ajesh, J. Nair, and P. S. Jijin, “A random forest approach for rating-based recommender system,” in International Conference on Advances in Computing, Communications and Informatics (ICACCI), Jaipur, India, pp. 1293-1297, 2016.

[2] S. Baccianella, A. Esuli, and F. Sebastiani, "Sentiwordnet 3.0: an enhanced lexical resource for sentiment analysis and opinion mining." pp. 2200-2204.

[3] Y. Bao, H. Fang, and J. Zhang, “Leveraging decomposed trust in probabilistic matrix factorization for effective recommendation,” in Proceedings of the 28th AAAI Conference on Artificial Intelligence (AAAI), Québec City, Québec, Canada, pp. 30-36 2014.

[4] D. Basak, S. Pal, and D. C. Patranabis, “Support vector regression,” Neural Information Processing-Letters and Reviews, vol. 11, no. 10, pp. 203-224, 2007.

[5] S. Berkovsky, and J. Freyne, “Group-based recipe recommendations: analysis of data aggregation strategies,” in Proceedings of the fourth ACM conference on Recommender systems, Barcelona, Spain, pp. 111-118, 2010.