預測Instagram平台上使用者對公司帳號之參與度

Predicting User Engagement of Business Accounts on Instagram

黃思凱(Szu-Kai Huang)

指導教授 ： 李瑞庭

國立臺灣大學/管理學院/資訊管理學研究所/碩士(2021年)

https://doi.org/10.6342/NTU202101490

摘要

許多公司透過它們在Instagram的商業帳號宣傳它們的產品或提升品牌能見度，這些帳號擁有相當多的追蹤者，但大部分的追蹤者很少參與公司的貼文或活動。越常參與公司活動的追蹤者，越可能幫公司進行相關產品或活動的口碑行銷，再則，公司間經常合作共同開發產品或舉辦相關活動，因此，幫這些公司找尋高參與度的追蹤者是迫切且不可或缺的議題。Instagram累積相當豐富的使用者產生內容，如：文字、照片與使用者互動等等，照片所包含的資訊可彌補充文字與使用者互動資訊的不足，但少有研究利用照片資訊預測使用者的參與度。因此，我們提出一個研究架構，從使用者產生的內容中擷取文字、照片與使用者互動特徵，然後利用這些特徵同時預測使用者對多家公司的參與度。實驗結果顯示，我們所提出的研究架構優於比較方法。我們的研究架構可幫助公司找出未來參與度高的使用者，讓公司增強與他們的互動，以提升行銷效率，進而幫助公司增加其客戶群、尋找潛在的忠實客戶以及擬定有效的行銷策略。

關鍵字

使用者參與度預測 ； 商業帳號 ； 社群平台 ； 深度學習 ； 注意力機制

並列摘要

Many companies promote their new products and brand images to potential customers through their Instagram accounts. These accounts usually accumulate a large number of followers. However, most followers are scarcely involved with account’s posts or activities. The more frequently the followers get engaged with company’s business account, the more likely they provide an arena for word-of-mouth marketing. Also, companies tend to cooperate with each other to develop a product or hold an event, it will be beneficial for them to find the followers who are simultaneously active in the business accounts of those companies. Therefore, it is desirable and essential to predict user engagement of multiple business accounts in the near future, where the user engagement reflects how frequently a user is involved with company’s posts. User-generated contents on Instagram contain various types of data such as text descriptions, photos, and user interactions. Photos contain rich information that can complement the information of text descriptions and user interactions. However, there is rare research using the features extracted from photos to predict user activeness or engagement. Therefore, we propose a framework to extract textual, photo and user interaction features from user-generated contents on Instagram, and then incorporate these features to predict user engagement. The experiment results show that the proposed framework outperforms the comparing methods in terms of mean absolute error. Our framework can help businesses identify highly engaged followers and maintain strong connections with them, which in turn help businesses expand their user base, find potential loyal customers and implement effective marketing strategies.

並列關鍵字

User engagement prediction ； business account ； social media platform ； deep learning ； attention mechanism

參考文獻

Blei DM, Ng AY, Jordan MI (2003) Latent Dirichlet Allocation. Journal of Machine Learning Research 3:993–1022.

Bottou L (2010) Large-scale machine learning with stochastic gradient descent. Proceedings of the International Conference on Computational Statistics. 177–186.

Chen J, Pirolli P (2012) Why you are more engaged: Factors influencing twitter engagement in occupy Wall Street. Proceedings of the International AAAI Conference on Web and Social Media. 423–426.

Chen R, Yang L, Goodison S, Sun Y (2020) Deep-learning approach to identifying cancer subtypes using high-dimensional genomic data. Bioinformatics 36(5):1476–1483.

Dadfarnia M, Matinpour AA, Abdoos M (2020) Churn prediction in payment terminals using RFM model and deep neural network. Proceedings of the International Conference on Information and Knowledge Technology. 8–10.