Research On Collaborative Conversation Recommendation Model Based On Deep Learning

In recent years,the development of Internet technology has advanced by leaps and bounds,and the scale and types of social information have become increasingly complex,and data cannot be processed in a timely and effective manner.This makes people face massive amounts of data in their lives.In this case,it is very difficult for users to choose suitable products and services from them.Therefore,the recommendation system came into being.Especially in the era of big data,the recommendation system plays an increasingly important role and has now become a basic tool for peoples’ work,study,entertainment,and life.With the help of recommender systems,people can obtain information more efficiently and make decisions more accurately.Traditional recommendation methods mostly rely on explicit preferences between users and items.However,in many practical cases,we cannot obtain enough information.For example,a user browses an online shopping store without logging in when shopping online.At this time,we cannot know the user’s historical behavior,and cannot obtain the user’s preferences.Only explicit observation information（such as clicks）is available.In order to solve such problems,scientists have proposed a session-based recommendation system.In recent years,researchers have proposed many methods based on conversational recommendation.With the vigorous development of deep learning,the conversation recommendation model based on recurrent neural network has achieved good results and has attracted widespread attention.Although current deep learning methods have achieved some good results,these methods have several problems,（1）long-term and short-term preferences of users are often considered separately,and（2）the modelling of existing methods focuses only on the current session in the current session,ignoring the synergistic information of neighbor sessions.This paper provides an in-depth study of session-based recommendation methods and proposes three effective methods.Firstly,a recurrent neural network-based model is proposed to address the problem of modelling users’ long-term and short-term preferences.A modified module of gated recurrent units is introduced which is interpretable and based on the theory of course learning allows for better learning ability.An attention network is also designed so that long-term and short-term preferences can be interacted with.Finally,to verify the effectiveness of the proposed network,an experimental environment is built and fully validated.The experimental results show that the proposed method improves on two real data sets and enhances the recommendation effect.Secondly,in order to capture information about items that are far away from each other within a session as well as the synergistic information provided by neighbor sessions,this paper combines a graph convolutional neural network with an attention mechanism.A key node is designed using the feature that graph convolutional neural networks can capture higher-order information for modelling purposes,and the attention mechanism is used to represent this key node by assigning corresponding weights to different items in the current session,while the synergistic information from neighbor sessions is taken into account in the updating process.The experimental results show that the proposed network model achieves better performance in terms of performance metrics.Finally,in several experiments,we found that graph convolutional neural networks generate huge parameters when modelling the whole session,as they take into account the items in the session,resulting in a large composition volume and a tendency to overfitting.This paper then proposes a collaborative attention model（CASR）based on recurrent neural networks.The model uses a traditional gated recurrent unit combined with an attention mechanism to extract features,while taking into account the collaborative information of neighbor sessions and putting it into a capsule network for updating.By comparing the same experimental environment,the model proposed in this paper achieves a large improvement in all performance metrics.