Modern systems of context advertisement do not always use analysis of user’s discussions and statements. If we analyze user’s expressions in the social networks, we can get an idea of their main topic, understand their attitude to some goods and then use all this information in the advertisement of some products and services. For example, user made a statement: “my mobile phone’s battery is poor”. Our system recognizes this comment’s purpose: it is about a mobile phone, and a user is not satisfied with it. In this case, social network’s advertisement algorithms can advise the user to purchase a new gadget. This system would increase CPA (click-per-action) rates and its conversion (a relation between advertisement cost and its revenue). As the result, it can help us gain some investors and advertisers, which will make the overall income grow.
The train dataset was collected from Reviws website.
Was used attention mechanism and toxic classification with BERT.
The acc is 82%
Alexandre Kalendarev Georgy Chernov