Research On Integrating Content Information Based Hybrid Recommender Systems

In the era of information explosion,recommender systems have become a bridge to match supply and need between information producers and consumers.The content of information is rich and diverse,which inspires us to integrate varieties of content information to improve the recommendation quality.Hybrid recommender systems can make full use of different information sources and take advantages of different types of recommender systems.We study integrating content information in hybrid recommender systems,and the main contributions are as follows:First,we improve content-based recommendation method.Conventional recommender systems exploit only explicit,singular modal rating feedback,the learned representations generally will be biased;and consequently,the recommendations are overspecified.To alleviate this problem,we propose DEAMER(Deep Exposure-Aware Multimodal contEnt-based Recommender),a model that combines multi-modal content information,jointly exploits rating and interaction signals via multi-task learning.DEAMER mimics the expose-evaluate process in recommender systems where an item is evaluated only if it is exposed to the user.DEAMER generates the exposure status by matching multi-modal user and item content features.Then the rating value is predicted based on the exposure status.We show that DEAMER outperforms state-of-the-art shallow and deep recommendation models on recommendation tasks.Second,we improve the explainability of recommender systems.Designing a persuasive product snippet generator can provide an explanation of recommended products for online sales,hence improving the success rate of recommender systems.However,lack of labeled data and subjective judgments pose severe challenges for making such a snippet generator.To tackle these issues,we design data-level,knowledge-level and model-level solutions and propose a persuasive product snippet generator called SILVER(SnIppet Loading Via intErest Relevance).Evaluations on real data demonstrate that SILVER is able to produce more fluent,catchy and informative snippets.Third,we alleviate the problem of obtaining large numbers of training samples.There are many system configurations in big data processing platforms(such as Spark),which must be set with different values to optimize workload performance.Machine learning methods can recommend more efficient configurations for different workloads.However,machine learning models usually need a large number of training samples.Since performing large workloads(i.e.,on large scale data set)is costly,it is unrealistic to obtain a great number of training samples.Therefore,we propose a new framework to collect training samples by running small workloads(which can collect more training samples by reducing the cost of a sample),and then predict the performance of big workloads.We verify the performance of many machine learning models in this recommendation scenario.The experimental results show that decision tree and ensemble model can recommend efficient configurations.