Unlocking insights: integrated text mining and interpretive structural modeling for enhanced user review analysis

Research Methodology

(1) Word frequency analysis. The former is the ratio of the number of occurrences of a specific word in the user reviews of milk products to the total number of words in the text of user reviews, and the latter is the ratio of the number of user reviews of all milk products to the number of studies containing a specific word, and finally, the words with the highest word frequency will be used as keywords in user reviews;

(2) Relevance analysis. Since TF-IDF cannot reflect contextual information, the keywords calculated by TF-IDF need to be analyzed for relevance, i.e., Word2vec is used to measure the similarity between words to obtain semantic relationships. Firstly, the expression vector of each keyword is calculated, according to which the spatial distribution of each word can be obtained, and then the related terms of each keyword can be found;

(3) Visualisation of the correlation situation. The correlation analysis can obtain the correlation between individual words but cannot know the spatial relationship of unique words in the overall review. Based on the logical connection between words and relevance, comments can be considered nodes and relevance as relationships. Based on this idea, the relevance of keywords can be visualized through the Neo4j graphical database, which can clearly and intuitively show the classification of the content of user comments and define different themes for each category containing other words according to the variety;

(4) Element relationship analysis. The different themes obtained through the data visualization are considered content elements reflecting the user’s needs, and the interrelationships and hierarchies between the components are explored using the ISM systematic analysis method.

Data acquisition

This experiment is based on Python 3.7, and the data is based on user reviews of milk products on Suning.com official website of Suning.com, and the target data is obtained through a self-written web crawler. Since the target website is a dynamically loaded web page, capturing the page information through the requests module is impossible, so it is necessary to use Selenium to automatically simulate the operation to crawl the HTML tags of the web page (Chen et al., 2021).

In this study, we crawled 60,367 comments from Yili, Mengniu, Junlebao, Gandhi Farm, Huang’s Dairy, Shengmu, Anjia, and Deya milk products on the official website of Suning.com, excluding invalid comments such as default comments, blank comments, and garbled text, and finally obtained 10,273 valid user comments.

Before formal text analysis, the text of the comments needed to be word-separated using jieba word-separation. To achieve more accurate word separation results and subsequent text analysis accuracy, punctuation and useless characters (numbers, tone words, etc.) were deactivated, i.e., these results were word filtered so as not to affect keyword extraction and word relevance analysis. The data for the first 50 user comments are shown in Table 1. The top 50 user comments on the milk product on the Suning.com website, as crawled by the crawler, reveal users’ different attitudes towards the product’s consumption.

Word frequency analysis

The jieba.analyze.extract_tags () function was used on the splitting results to obtain the keywords in the comment text, with the top K parameter set to 20, indicating that the top 20 keywords with the most significant TF-IDF weight values were returned (Budel et al., 2023), and the results are shown in Table 2. As seen from Table 2, users’ concerns about milk products can be roughly divided into elements, including taste, freshness, logistics and price, which are also user needs for food products. Based on the TF-IDF values in Table 2, the keywords in the results of this word frequency analysis were output as a word cloud diagram (Fig. 2).

The word cloud diagram in Fig. 2 clearly shows the frequency of keywords, especially those that appear most frequently in the comments. For example, “delicious” and “taste” indicate that the main focus of customers on milk products is on the taste of the product itself, while “date” and “fresh” point to the freshness of the product, while “price”, “cheap” and “affordable” point to the price and benefits of the product itself. These are the core content that users are most concerned about, so e-commerce platforms or manufacturers can use the word cloud to dig deeper into users’ demands and ideas and also to give users a general understanding of their products so that they can quickly refine product information before making their own purchasing decisions.

Relevance analysis

To obtain the interrelationships between the keywords, i.e., to get the relevant words for each keyword, Word2vec was used for the relevance analysis. In this section, the Word2vec model is invoked for training, with default values used for most of the model parameters and other parameters set (Table 3).

As shown in Table 4, the word vector of each word shows its position in space, and the phrases most closely related to each keyword can be analyzed according to their distance, i.e., the terms related to each keyword can be calculated from the word vector (Heimerl & Gleicher, 2018).

Table 5 indicates the degree of relationship with the keyword-related words, and the higher the value, the closer the relationship with the word, i.e., closer to the meaning or pointing content of the keyword (Benito-Santos & Sanchez, 2019; Radhakrishnan et al., 2017). The related words for each keyword are shown in it. For example, the associated terms for the keyword good to drink are very fragrant, fragrant, fabulous, indeed, sour, and sweet, where the most relevant observation is very aromatic, reaching 0.7561, indicating that users think milk is good to drink mainly from the fragrance of milk (Fig. 3). To show the relationship situation of all words more visually, the relevance situation of all keywords was visualized by drawing a knowledge graph through Neo4j.

According to Table 6, classification of the results of the relevance analysis of user comments. Based on the results of the keyword relevance analysis in Fig. 3, it can be seen that the user comments on the items are divided into eight sections.

Analysis of influencing factors

By analyzing Table 6, it can be seen that Category 3 and Category 5 both described content in terms of logistics and express delivery and, therefore, could be classified into one category; while Category 4 and Category 7 both referred to the situation where consumers trusted the shop and thus returned to it repeatedly, and therefore could also be classified into one category. The above analysis was combined, and Table 6 was rearranged to finally organize the user review content into six categories and group the categories into different themes based on the range of the keywords included (Gonalves et al., 2023), with the complete results shown in Table 7. The six subject terms collated as the six elements of user comments are analysed by the ISM explanatory structural model to investigate the influence between the elements.

As shown in Table 8, drawing on the findings of Stan et al. (2010) and others, the following can be specified: freshness of taste and discounted prices offer both influence customer repurchase, while customer repurchase, logistics and product packaging all influence discounted prices of the product. In addition, discounted price and product packaging affect logistics, freshness of taste directly affects nutritional composition. this shows the adjacency matrix of the model, the initial input matrix, abstracted from the relationships between the elements.

According to Table 9, the adjacency matrix multiplication of the model, which is an intermediate computational procedure to obtain the reachable matrix by successively multiplying the adjacency multiplication matrix until the matrix does not change (Mallick, Chaudhari & Joshi, 2022).

As shown in Table 10, with the adjacency matrix representing the direct relationship between elements and the accessibility matrix representing whether the transmission between elements leads to an indirect influence relationship.

Table 11 shows the hierarchical decomposition process of the model. At each extraction, if the intersection of the reachable and reachable sets with the initial stage is the same, the element set can be extracted for hierarchical decomposition (Fernandes, Simsek & Kantarci, 2021). Table 12 shows the hierarchy of models, where the top level is the system’s ultimate goal, while the following levels are the reasons for the upper levels respectively.

A directed connectivity diagram based on the results of the hierarchy can show the scale of the model more visually (Fig. 4).

User review results analysis and route optimisation

Analysis of results

By analyzing the user reviews of milk products on Suning.com, the data were mined, analyzed, and comprehensively evaluated through three methods, TF-IDF, Word2vec, and ISM, in turn. The above three methods could drill a large amount of characteristic information, interrelationships, and the influence between each element from the user reviews, thus e-commerce platforms, producers, and other system management inspiration and decision-making advice (Bhatt & Kankanhalli, 2011).

(1) Through the overall analysis of user reviews, its main content revolves around six elements of milk products: freshness of taste, discounted prices, logistics, customer repurchase, product packaging, nutritional composition;

(2) TF-IDF was able to dig out the words that appeared more frequently in the user reviews. In the word frequency ranking, “delicious” and “date” occupy the top positions in word frequency. In the top 10 rankings, keywords related to taste and texture ranked 1st, 3rd, and 5th, respectively. In contrast, keywords related to freshness ranked 2nd and 4th, respectively, indicating that users focus on the taste and freshness of the milk product. In addition, the 7th and 9th rankings are for “logistics” and “shipping”, indicating that users are particularly concerned about the logistics of the product, in addition to the taste and freshness of the product;

(3) Word2vec can analyze the related words of each keyword to obtain the connection between each word. For example, “date” is related to “freshness”, “shelf life“, “date of production” and so on. “price” is similar to “prices”, “event”, and “economical”;

(4) The knowledge graph drawn by Neo4j can visualize the relevance of each word, visually display the connection between all comments, and categorize the overall content of user reviews into specific categories according to the classification, defining different topics according to the class. For example, words such as “delicious”, “milk quality”, “taste” and their related words point to taste the freshness, “regular clients”, “repurchase”, and “long-term”, and their associated terms point to customer repurchase;

(5) ISM can analyze the interplay between the elements. In tier 3 elements, taste freshness affects discounted prices, customer repurchase, and nutritional composition; in tier 2 elements, product packaging affects discounted prices, and logistics, nutritional composition affects customer repurchase; in tier 1 elements, discounted prices and customer repurchase affect each other and logistics, while price offer involves logistics and logistics express affects customer repurchase. Since milk is a nutritious food that can be consumed directly, the results of this study can be applied to other beverages, fruits and vegetables, and other foods. For manufacturers, the results can be used to improve their products, such as strictly controlling the packaging and paying attention to the degree of freshness of the food, etc. For consumers, it is important to pay attention to the time to consume the products in time to prevent them from exceeding the shelf-life.

Path optimisation

According to the hierarchy diagram in Fig. 4, the elements of layers 3 and 2 points to the producers, while the aspects of layer one end to the e-commerce platform, i.e., the e-commerce platform is the core of the entire hierarchy. Also, it serves as a bridge between the producers and consumers, whose goal is to attract consumers and promote product sales by providing convenient shopping channels and an engaging shopping experience. The smooth operation of an e-commerce platform depends on its operational strategy and the quality of the products supplied by producers (Kolay, 2015).

For producers, ensuring the freshness of their dairy products means adopting appropriate production and storage methods to ensure that products remain fresh and of high-quality during transport and distribution and that hygiene standards and quality requirements are met at all stages. At the same time, producers should ensure that milk products have an excellent nutritional composition and meet the health needs of consumers by adding essential elements and other nutrients to the ingredients following scientific formulation requirements to provide consumers with a balanced choice of products. In addition, manufacturers should also focus on product packaging design. Good packaging not only attracts consumers’ attention and communicates the quality and value of the product to them but also provides the necessary protection to ensure that the product is not damaged during transportation and delivery.

When upstream producers can deliver high-quality products, downstream e-commerce platforms can develop appropriate and sensible marketing strategies to attract purchases and repeat purchases. For example, e-commerce platforms can highlight a product’s freshness and nutritional value and increase its exposure and appeal by showcasing its benefits and detailed descriptions. In addition, e-commerce platforms can promote sustained purchase behavior by actively collecting user feedback and reviews to improve product quality and service and continuously increase user satisfaction.

There is an interdependent relationship between the e-commerce platform and the manufacturer. The manufacturer’s provision of high-quality products is a prerequisite for the smooth operation of the e-commerce platform. In contrast, the effective marketing strategy of the e-commerce platform can facilitate the sale of the manufacturer’s products and user repurchases. Through close cooperation and continuous product quality improvement, both parties have jointly promoted the successful operation of milk products on the e-commerce channel.