Efficient parallelization for big data collaborative recommendation decisions

This paper proposes a novel Sentiment-Based Probabilistic Tensor Analysis technique {senti-PTF} to address the information overload problem through information filtering. The proposed framework first applies a Natural Language Processing (NLP) technique to perform sentiment analysis taking advantage of the huge sums of textual data generated in from the social media are predominantly left untouched. Although some current studies do employ review texts, many of them do not consider how sentiments in reviews influence recommendation algorithm for prediction. Existing works concentrate only on rating matrix which are often sparse. There is therefore this big data text analytics gap whose modeling is computationally expensive. Probabilistic Tensor Factorization (PTF) is a standard technique for such large scale processing. From our experiments, our novel machine learning sentiment-based probabilistic tensor analysis (senti-PTF) is computationally less expensive, scalable and addresses the scalability problem and cold-start problems, for optimal recommendation decision making as shown from our error detection evaluation metrics.