| Recommendation algorithms have facilitated technological advancements in various fields,providing new opportunities for their development.Nowadays,there is a growing demand for personalized recommendations in different industries,including the inspection and testing field.However,applying recommendation algorithms to specific fields requires building a data system from scratch,heavily relying on domain experts to analyze the specific characteristics of the field.The objective of this research is to apply recommendation algorithms in the field of inspection and testing,focusing on the data characteristics of user-project interactions.By introducing a user multi-behavior recommendation algorithm,this article aims to address the strong field-specific limitations,sparse data,and cold-start issues that exist in the application of inspection and testing.Therefore,this study carries out research on inspection and testing recommendation models,with the main work including:(1)The multi-behavior recommendation models emphasize capturing the differences between various interaction behaviors,but overlook the commonalities between them.This paper proposes a novel Multi-Behavior Heterogeneous Contrastive learning Recommendation(MBHCR)model based on graph convolutional networks,which focuses on fusing information between different interaction behaviors and capturing their interaction differences on a behavior-view level.In this work,MBHCR models all types of behavior interactions as a unified heterogeneous graph to represent behavior information and aggregates the relationships between user-project behavior interactions through a multi-behavior relationship aggregator.Then,through behavior contrastive learning enhancers,it captures the interaction differences between different behavior information.Finally,experiments on two real-world datasets demonstrate the effectiveness of MBHCR.(2)Multi-behavior interactions provide more comprehensive and complex interaction information for recommendation models.However,current methods often model the relationships between different user interactions by controlling many hyperparameters,leading to the problem of over-parameterization of the model.This paper proposes a new Variational Auto-Encoder Heterogeneous Graph Multi-Behavior Recommendation(V-GMR)framework based on multi-behavior heterogeneous graph for efficient capturing of user behavior preferences while addressing the problems.Firstly,V-GMR introduces Variational Auto-Encoder to learn features using feature encoding,to capture the feature representation of multi-behavior information well.In addition,an information fusion layer is further created to integrate target behavior with latent auxiliary preference features.Finally,this paper verifies the proposed V-GMR on three datasets,and the experimental results show that V-GMR outperforms the latest baselines in terms of recommendation performance.(3)To apply recommendation algorithms in the field of inspection and testing,the first step is to collect,clean,and organize multi-behavior user-item interaction information in the field of inspection and testing.This data can be stored using a Mysql database system.Through experimental comparison,we applied the designed V-GMR model to the inspection and testing recommendation system and designed a visualization interface to provide users with efficient project recommendation results based on the application and functional requirements of the inspection and testing field.The experimental results show that the two multi-behavior recommendation models proposed in this paper perform well on public datasets,and they also demonstrate targeted and efficient performance when applied to recommendation in the field of inspection and testing. |
