Weakly-supervised consumption intent detection in microblogs

Nowadays, more and more people are using microblogs to discuss various topics, sometimes revealing their potential consumption intent (CI). A consumption intent is a desire or hope for something to purchase. This paper proposes a weakly-supervised approach to detect those user posts containing consumption intent in microblogs. First, we collect a large number of posts containing a few designated hashtags and regard them as containing CI (positive instances). Posts without such hashtags are considered as negative instances. Then, we train a statistical classification model using the training data constructed as above and design three kinds of post-dependent features. We evaluate the model on 7, 902 manually annotated posts from microblogs and show that the proposed method is effective, even outperforms the supervised method.