Main Article Content
In this paper, we aim to enhance the relevance of e-commerce web sites by a prior quality checking of proposed products. This checking is done by analyzing social networks and YouTube videos comments.
To achieve this objective, we have broken down the work into a few steps. The first one consists in scraping text from social networks groups and storing it in a NoSQL graph database. Each scraped word is linked to one or many reactions that are coming from the social network. Therefore, we can utilize the database as a knowledge source that associates each set of terms with a specific type of reaction: positive, negative or neutral. Afterwards, we use a TF-IDF based filtering method to keep only relevant words and eliminate those which are a connected to all reactions. The advantage of this stage is the presence of a knowledge source that can be used for product quality checking.
In the e-commerce web site, data are coming from multiple e-commerce websites. The latter, offer products without quality checking. Concretely, we aim to allow users to check quality by a simple check button which call an implemented web service using human reactions and comments. After evaluating our approach, we have obtained an accuracy of 0,75. This result means that our method gives a three quarter of chance to have a good product.