【作者】 刘平；

【导师】 黄虎杰；

【作者基本信息】 哈尔滨工业大学 ， 软件工程， 2021， 硕士

【摘要】 随着我国电子商务的迅速且稳步的发展,网上购物商品种类分类越发精细,物流配送愈发方便,消费者更加愿意进行网上购物消费。人们逐渐抛弃传统的大型实体购物商店消费方式,转向新型网络购物商店。消费者购买并使用了某个产品后,就可能会对商品进行用户评分和商品评论来表达这次购物的体验。同时,人们在购买商品时,往往会先去查看产品评论的信息,对商品做一个了解。用户评价通常包括一个数字评分和文字评论,这些信息内容反映了用户对不同的产品属性特点和用户情感倾向的偏好。推荐系统在电子商业中得到了广泛的研究与应用。各电商平台都提供"猜你喜爱"的选择功能。现阶段,人们不缺少收集信息的渠道,反而如何在信息过载的情况下进行个性化推荐成为了关键,推荐系统其核心就是通过一种个性化算法,利用用户对不同的商品所做出的评价信息,挖掘出其兴趣偏好本文以BERT预训练的语言学习模型、BLSTM网络模型以及Attention注意力驱动机制模型结合构建一种神经网络模型,对用户评论文本进行偏好分析。首先构建网络爬虫工具收集商品页面的用户评论信息,预处理文本数据并将其存入可操作数据库中。通过BERT对评论文本进行预训练,然后采用BLSTM提取双向上下文特征,接着使用注意力机制对信息权重进行分配。获得用户评论关注点,而后针对用户兴趣偏好模型对不同用户生成个性化推荐,推荐效果比传统的推荐系统要更精确。更多还原

【Abstract】 With the rapid and steady development of e-commerce in my country,the types of online shopping products are becoming more and more refined,logistics and distribution are becoming more convenient,and consumers are more willing to make online shopping.People gradually abandon the traditional large-scale physical shopping store consumption mode and turn to new online shopping stores.After consumers purchase and use a product,they will make user ratings and product reviews on the product to express their shopping experience.At the same time,when people make purchases,they often first check the product review information to get an understanding of the product.User reviews usually include a digital score and text comments,these information content reflects the user’s preference for different product attributes and user emotional tendencies.Recommendation systems have been widely studied and applied in electronic commerce.All e-commerce platforms provide the option of "Guess what you like".At this stage,people do not lack channels for collecting information.Instead,how to make personalized recommendations under the circumstances of information overload has become the key.The core of the recommendation system is to use a personalized algorithm to use the user’s evaluation of different products to dig out their interests and preferences.This paper uses BERT pre-trained language learning model,BLSTM network model and Attention drive mechanism model to construct a deep learning model to analyze the preference of user comment text.First,build a web crawler tool to collect user comment information on product pages,preprocess the text data and store it in an operational database.Then use BERT to pre-train the comment text,use BLSTM to extract two-way context features,and then use the attention mechanism to assign information weights.Obtain user comment attention points,and then generate personalized recommendations for different users based on the user’s interest preference model.The recommendation effect is more accurate than traditional recommendation systems.更多还原