There is a high correlation between user behavior and user features in recommender systems. User review texts reflect user preferences and item feature information. However, the current research on CTR prediction models based on user behaviors fails to fully mine user features. As a result, the prediction accuracy of the model is not high. To solve this problem, we propose a click-through rate prediction model that fuses user comment text and behavior sequence. The model uses a text convolutional neural network to extract the features of user review text to obtain the feature vector of user comment text, and uses an attention mechanism to capture the user's interest points from the user's behavior sequence to obtain the user's interest feature vector. A multi-layer perceptron is then used to fuse the user's comment text feature vector, interest feature vector and item feature vector for click-through rate prediction. The experimental results show that the proposed model has better prediction performance than current click-through rate prediction models.
How to automatically obtain cross-features with different weight values is a significant issue in the research of recommendation models. Traditional recommendation models cannot automatically learn the deep-level features of users and items to obtain cross-features. The mixed processing of dense numerical features and sparse categorical features will result in more information loss during dimensionality reduction. Cross features occupy the same weight in the recommendation process, which will lead to the non-prominence of critical features and reduce the accuracy of model recommendations. This paper proposes a personalized recommendation model (MSRN) for self-attention perceptron with automatic feature correlation. The model first processes the numerical features and category features in double towers to reduce the loss of feature information. Numerical cross-feature matrix and category cross-feature matrix use multilayer perceptrons to automatically mine the hidden knowledge and relationships between features. The model uses the Hadamard product to process it to obtain the cross feature matrix and uses the self-attention mechanism to assign different weights to the extracted cross-features. The experimental results on the public data set show that the recommended evaluation indicators of this model, MAE, and RMSE, are better than the current advanced recommendation models and have better accuracy and stability.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
INSTITUTIONAL Select your institution to access the SPIE Digital Library.
PERSONAL Sign in with your SPIE account to access your personal subscriptions or to use specific features such as save to my library, sign up for alerts, save searches, etc.