信息工程与人工智能学院知识库

近年来，随着电子商务的快速发展，网上购买农产品作为购物的方式之一，深受消费者喜爱。然而，由于电商平台的农产品种类繁多，用户需要花费更多的时间和精力去寻找自己喜欢的产品，降低了购物效率，影响了购物体验。因此，对于电商平台来说，依据消费者的喜好，针对性的推荐就显得非常重要。传统的农产品推荐大多基于用户对所购买产品的评分推断用户兴趣，进而进行推荐，却忽略了产品在线评论对推荐效果的影响。而在线评论蕴含着大量用户对产品特征个人喜好方面的信息，对个性化推荐极为重要。鉴于此，本文通过对国内知名电商平台的在线评论进行分析，以甘肃省特色农产品为研究对象，构建了基于情感分析的个性化推荐模型。该模型在情感分析基础上，利用矩阵分解、改进巴氏系数相似度以及混合协同过滤算法对传统协同过滤算法进行了优化，进而提升推荐的准确性。概括起来，本文的主要工作如下：

（1）拓展了农产品领域的情感词典，并基于情感词典计算用户评论的情感值，得出情感评分矩阵。在现有情感词典的基础上，利用情感倾向点互信息SO-PMI算法和LDA主题模型对农产品领域的情感词典进行拓展，并加入网络词词典进行完善，以此计算出用户评论的情感值，从而得出情感评分矩阵。

（2）构建了融合矩阵分解和改进巴氏系数的混合推荐算法。首先，利用基于偏置的矩阵分解BMF算法对评分矩阵进行缺失值填充；其次，对传统巴氏系数相似度度量方法进行改进；再次，对基于用户和基于项目的两种协同过滤推荐算法进行融合；最后，将BMF算法、改进巴氏系数以及混合协同过滤算法融合得到混合推荐算法，并在数据集上验证了算法的准确性。

（3）构建了基于情感分析的农产品个性化推荐模型。基于词典的情感分析得到情感评分矩阵，在此基础上，融合混合推荐算法，建立基于情感分析的个性化推荐模型，并与UCF、ICF、HCF三种推荐模型对比，验证模型的有效性。

（4）设计了农产品个性化推荐系统。将基于情感分析的个性化推荐模型应用在推荐系统中，从而为用户提供个性化的农产品推荐。

In recent years, with the rapid development of e-commerce, buying agricultural products online has been deeply loved by consumers as a way of shopping. However, due to the wide variety of agricultural products on e-commerce platforms, users need to spend more time and energy to find their favorite products, which affects the shopping experience. Therefore, for e-commerce platforms, targeted recommendations are very important according to consumers' preferences. Traditional agricultural product recommendation is mostly based on the user’s rating of the purchased product to infer user interest, thereby making recommendations, but ignores the impact of product online reviews on the recommendation effect. Online reviews contain a large number of users' personal preferences for product features, which are extremely important for personalized recommendations. In view of this, this article analyzes the online reviews of well-known domestic e-commerce platforms, takes Gansu Province's characteristic agricultural products as the research object, and constructs a personalized recommendation model based on sentiment analysis. The model is based on sentiment analysis, using matrix decomposition, improved Bhattacharyya coefficient similarity and hybrid collaborative filtering algorithm to improve the traditional collaborative filtering algorithm, and then improve the accuracy of recommendation. In summary, the main work of this article is as follows:

(1) The sentiment dictionary in the field of agricultural products is expanded, and the sentiment score matrix of user reviews is calculated based on the sentiment dictionary. On the basis of the existing sentiment dictionary, the sentiment dictionary in the field of agricultural products is expanded by using the sentiment point mutual information algorithm and the LDA topic model, and the network word dictionary is added to improve it, so as to calculate the sentiment value of the user's comment, and then obtain sentiment score matrix.

(2) A hybrid recommendation algorithm combining matrix decomposition and improved Bhattacharyya coefficient is constructed. Firstly, use the bias-based matrix factorization algorithm BMF to fill in the missing values of the score matrix; secondly, improve the traditional Bhattacharyya coefficient similarity measurement method; thirdly, integrate the two collaborative filtering recommendation algorithms based on user and item ; Finally, the BMF algorithm, improved Bhattacharyya coefficient and hybrid collaborative filtering algorithm are combined to obtain a hybrid recommendation algorithm, and the accuracy of the algorithm is verified on the MovieLens data set.

(3) A personalized recommendation model of agricultural products based on sentiment analysis is constructed. Based on the sentiment analysis of the dictionary, the sentiment score matrix is obtained. On this basis, a hybrid recommendation algorithm is combined to establish a personalized recommendation model based on sentiment analysis, and the superiority of the model is verified by comparison with UCF, ICF, and HCF three recommendation models.

（4）A personalized recommendation system for agricultural products is designed. The personalized recommendation model based on sentiment analysis is used in the recommendation system to provide users with personalized recommendation of agricultural products.