TAFM: A Recommendation Algorithm Based on Text-Attention Factorization Mechanism

Figure 1

The overall pipeline of the proposed TAFM. The user attribute data go through the preprocessing layer and transform from sparse vectors into dense vectors. The dense vectors are computed in parallel in the FM, deep, intent understanding, attention, and autoencoder layers. Finally, the results of each layer are summed, sigmoid computation is performed, and the text data are fed directly to the intent mining layer in the deep layer.