A Self-Attention Bidirectional Long Short-Term Memory for Cold Start Movie Recommendation Models
M. Manohar, R Archana Reddy, Laith H. Jasim Alzubaidi, Abbas Hameed Abdul Hussein, P. R. Buvaneswari · 2023
Movie recommendation systems are useful tools that help users find relevant results and prevent information overload. On the other hand, the user cold-start issue has arisen because the system lacks sufficient user data. Furthermore, they are not very scalable for use in extensive real-world applications. One of the key strategies to address the sparsity and cold-start problems is to leverage other sources of information, including item or user profiles or user reviews. Processing client feedback is typically a challenging process that involves challenging the interpretation and analysis of the textual data. Thus, this research implements an efficient deep learning-based recommendation architecture. Following the acquisition of textual data from the Amazon product reviews database, stop word removal, lemmatization, and stemming techniques are applied to the data pre-processing which eliminate inconsistent and redundant data, facilitating the process of interpreting and utilising data. Then, the Term Frequency-Inverse Document Frequency (TF-IDF) method is applied to extract the feature values from the pre-processed text data. The extracted feature values are fed to the Self-Attention Bidirectional Long Short-Term Memory (SA-BiLSTM) that utilises the matrix factorization method framework's information sources. The SA-BiLSTM model obtained 95.93% of recall, 94.76% of precision, and 97.84% of accuracy on the amazon product reviews database.