Microblogging Reviews Based Cross-Lingual Sentimental Classification for Cold-Start Product Recommendation

Nowadays many e-commerce websites allow users to sign in to their websites through user's social media account like twitter, facebook. This is basically for acceptance of e-commerce on social networks. This lead to failure of current recommendation systems to recommend items to users who sign in using social media account on e-commerce website. This problem commonly known as cross website cold start product recommendation. In cold start situation either product or user is new and recommendation system don't have any previous information about them and hence fails to recommendation of items. Our system addresses this issue using microblogging reviews of items collected from social media platform, Twitter. Both audio or video reviews and text reviews in English and Hindi languages are considered in recommendation. Speech to text processing is done on audio or video reviews to get sentimental classification. Cross Lingual Sentiment/Topic Model is used to carry out sentimental classification of Hindi languages reviews. Mapping of product's features and user's features is done with the matrix factorization method. Our experimental results on dataset prepared from microblogging platform Twitter and e-commerce website Amazon demonstrate that our model significantly improve the accuracy of recommendation system in cold start situation.