Research On Recommendation Algorithm And System Based On Deep Learning

With the development of communication technology and the Internet industry,we are currently not only in the information age,but also in the age of "information overload”.Whether it is online APP or various offline platforms,the amount of data generated is growing exponentially.In a wide variety of products,constantly updated news,and overwhelming advertisements,the information generated far exceeds our personal information acceptance,and it is difficult for us to obtain information of our own interest.However,with the in-depth application of information technology and big data technology,the development of personalized recommendation technology and recommendation system has also been continuously promoted.At present,various e-commerce,video,knowledge content and other platforms,independent design and research of recommendation systems are becoming more mature,but in the Internet medical field,effective personalized recommendation applications are still lacking.Therefore,this paper proposes a recommendation system based on deep learning with enhanced recall and ranking,which aims to solve the problem of article recommendation in the online medical field.This article takes the recommendation model based on deep learning as the main research content,and aims to make medical article recommendation more accurate and effective through the research and application of the recommendation model in this article.Based on the recommendation model of enhanced recall and attention mechanism,in response to the problem of poor recall effect when single-way recall is used in traditional recommendation,an enhanced recall strategy is proposed to recall a large number of articles,and deep learning technology is used to deepen the neural network.Based on the network,the attention mechanism is introduced to allow the user's historical click behavior to be taken into account when recommending articles,so that the recommended articles are more in line with user characteristics.The thesis mainly completes the following three aspects of work:(1)According to the composition of the recommender system and the division of each functional stage,an enhanced recall strategy is proposed in the recall stage of the recommender system,and articles are recalled in a variety of ways,and the results of various recalls are merged as candidate articles,which improves the recall effect and provides for the ranking stage good data foundation.(2)In the ranking stage of the recommendation system,candidate articles are scored and their click rates are predicted.Feature analysis of articles is required.Traditional models can usually solve one-dimensional or two-dimensional feature intersection problems,but in high-dimensional feature intersection analysis However,it cannot be solved effectively.Therefore,this paper uses deep neural networks to learn the hidden information in the process of high-dimensional feature crossover through neural networks,and introduces an attention mechanism to enable effective interaction between the list to be recommended and the behavior of users.(3)Based on the model of the text research design,in order to explore the recall effect and recommendation effect of the model in medical article recommendation,a control experiment of the recall phase and the overall model is carried out.Using Tianchi open source data,the data set is the click log data of medical articles of 300,000 users of a mobile application.The control experiment is divided into single-channel recall comparison experiment,enhanced recall comparison experiment and baseline model comparison experiment,through multi-dimensional comparison to explore the recommended effect of the research model in this paper.The experimental results show that the model proposed in this paper has an excellent effect on the recommendation of medical articles.The article recall evaluation reaches 0.2111,and the final recommendation evaluation reaches 0.2909,which is better than other comparison models.