Search optimization method and system for procurement platform based on intelligent recommendation algorithm

Abstract

This application relates to the field of search optimization technology, specifically to a search optimization method and system for a procurement platform based on intelligent recommendation algorithms. The method includes: evaluating the consistency of user behavior between two search terms; calculating the behavioral correlation between any two search terms; mining search terms that are synonymous with the current user's search terms; constructing a search extension set; and selecting products whose product information contains search terms from the search extension set as candidate products; determining the matching coefficient for each candidate product; determining the semantic correlation and correlation coefficient between any two products based on all users' historical purchase records; and determining the priority recommendation value for each candidate product based on the correlation coefficients between different products and each candidate product in the current user's historical purchase records, combined with the matching coefficients, thereby generating a recommendation result list for intelligent recommendation and display. This application significantly improves user search efficiency, procurement experience, and platform cross-selling conversion rate.

Description

Technical Field

[0001] This application relates to the field of search optimization technology, specifically to a search optimization method and system for a procurement platform based on intelligent recommendation algorithms. Background Technology

[0002] With the rapid development of e-commerce, the number of products on procurement platforms has grown exponentially. How to help users quickly and accurately find the products they need from a massive amount of goods has become a core issue in improving user experience and platform conversion rates.

[0003] Traditional methods for searching using search terms typically employ simple text matching techniques. This involves comparing the user's input search terms with product information to return search results. Since user-input search terms often contain numerous synonyms or near-synonyms, relying solely on text matching easily leads to the omission of many truly relevant products, resulting in incomplete search results. Secondly, traditional methods ignore the personalized preferences implied by users' past purchase records and browsing habits, lacking personalized display of search results. Furthermore, they fail to uncover the inherent connections between products, making it difficult to capture users' potential purchasing needs, reducing the efficiency of cross-selling and recommendations. Consequently, recommended products cannot accurately match users' true purchasing intentions, resulting in low search conversion rates and a poor overall search experience. Summary of the Invention

[0004] To address the aforementioned technical issues, a search optimization method and system for procurement platforms based on intelligent recommendation algorithms are provided to resolve existing problems.

[0005] The solution to the technical problem in this application is to provide a search optimization method and system for a procurement platform based on intelligent recommendation algorithms, including the following steps: In a first aspect, embodiments of this application provide a search optimization method for a procurement platform based on an intelligent recommendation algorithm, the method comprising the following steps: Extract product information and historical search logs recorded on the platform. Based on the number of clicks and browsing time of products under any two search terms in the logs, as well as the overlap of clicked and browsed products, evaluate the consistency of user behavior between the two search terms, calculate the behavioral correlation between any two search terms, and thereby mine search terms that are synonymous with the current user's search terms. Construct a search extension set, and filter out products whose product information contains search terms within the search extension set as candidate products. By retrieving the frequency and distribution of search terms within the extended set in the product information of the candidate products, the matching coefficient representing the user's search intent for each candidate product is determined. Based on the historical purchase records of all users, the semantic correlation of any two products at the content level is determined according to the keyword relevance contained in the product information between any two products and the behavioral correlation when the keywords are searched. The correlation coefficient of any two products is obtained by combining the frequency of the two products being purchased by users at the same time. Based on the correlation coefficients between different products and each candidate product in the current user's historical purchase records, combined with the matching coefficients, the priority recommendation value of each candidate product is determined, and a recommendation result list is generated for intelligent recommendation and display.

[0006] Preferably, the calculation of the behavioral relevance between any two search terms includes: Traverse each search term in the historical search log, obtain all products clicked by the user under the search term, and form a set of clicked products; find the intersection of the clicked product sets of any two search terms, analyze the differences in click and browsing behavior of the same products under the two search terms within the intersection, and determine the degree of behavioral difference between the two search terms. Calculate the relevance between the clicked product sets of any two search terms, and use this as the product overlap. The behavioral correlation is positively correlated with the product overlap, but negatively correlated with the behavioral difference.

[0007] Preferably, the calculation process of the behavioral difference is as follows: calculate the difference in the number of clicks of each product in the intersection under any two search terms, and record it as the number difference; calculate the average browsing time of each product in the intersection under any two search terms, and record it as the average duration; the behavioral difference is positively correlated with both the number difference and the average duration.

[0008] Preferably, the process of constructing the search extension set is as follows: for each search term, the remaining search terms whose relevance to its behavior is greater than or equal to a preset threshold are defined as synonyms, and the current user's search terms and all its synonyms are combined to form a search extension set.

[0009] Preferably, the matching coefficient is calculated as follows: The product information includes product title text and product details text; the total number of times all search terms in the search extension set appear in the product title text of each candidate product is counted as the first match degree of each candidate product; The total number of times all search terms within the search extension set appear in the product detail text of each candidate product is recorded as the term frequency factor. The product detail text of each candidate product is divided into sentences, and the number of sentences containing any search term within the search extension set is counted. The proportion of each sentence in the total sentences of the candidate product is calculated and recorded as the sentence coverage rate. The product of the sentence coverage rate and the term frequency factor is used as the second matching degree of each candidate product. The matching coefficient is the result of a positive fusion of the first matching degree and the second matching degree.

[0010] Preferably, the process for determining the semantic relevance is as follows: For each product, the product information is segmented into words, all keywords are extracted, and the word vectors of each keyword are obtained. For all users' historical purchase records, select one keyword from each of the two products and define them as a keyword group; calculate the correlation between the word vectors of the two keywords in the keyword group, and record it as semantic similarity; obtain the behavioral relevance when using the two keywords in the keyword group as search terms; positively fuse semantic similarity and behavioral relevance as the relevance contribution value of the keyword group; For any two products, for any keyword of one product, select the maximum association contribution value of the keyword group formed by that keyword and all keywords of the other product; The semantic correlation between any two products is positively correlated with the maximum correlation contribution value.

[0011] Preferably, obtaining the correlation coefficient between any two products includes: calculating the percentage of users who simultaneously purchased any two products based on all users' historical purchase records, as the product co-occurrence rate; the correlation coefficient is positively correlated with both semantic correlation and product co-occurrence rate.

[0012] Preferably, determining the priority recommendation value for each candidate product includes: For each item purchased in the current user's historical purchase record, obtain its correlation coefficient with each candidate item; and positively fuse the correlation coefficients between each candidate item and all items purchased by the current user as a personalization factor for each candidate item. The priority recommendation value is positively correlated with the matching coefficient and the personalization factor.

[0013] Preferably, the process of obtaining the recommendation result list is as follows: for the current user's search terms, all candidate products are sorted in descending order according to their priority recommendation value to generate a recommendation result list.

[0014] Secondly, embodiments of this application also provide a procurement platform search optimization system based on an intelligent recommendation algorithm, including a memory, a processor, and a computer program stored in the memory and running on the processor. When the processor executes the computer program, it implements the steps of any of the above-described procurement platform search optimization methods based on intelligent recommendation algorithms.

[0015] This application has at least the following beneficial effects: This application calculates the behavioral relevance of any two search terms. Its beneficial effect lies in analyzing historical user clicks and browsing behavior on products under different search terms during the search process, accurately identifying synonymous search terms from a behavioral perspective. Based on this, it mines search terms that are synonymous with the current user's search terms, constructs a search extension set, and filters out products containing search terms from the search extension set as candidate products. Its beneficial effect is that it breaks through the limitations of traditional literal matching; using the search extension set for product recall can significantly expand the recall scope, effectively solving the problem of missed product recall due to different user input habits, and improving the comprehensiveness of search results. Determining the matching coefficient of each candidate product has the beneficial effect of not only considering the frequency of search terms in the search extension set in product information but also deeply analyzing their distribution in product information, measuring the fit between the information content of candidate products and the user's search intent from multiple dimensions, and more accurately identifying products that best match the user's core query intent. Determining the semantic relevance of any two products has the beneficial effect of considering historical purchase records. The semantic relevance of two products at the content level is analyzed to uncover their inherent connections in terms of function, attributes, or application scenarios. The correlation coefficient between any two products is obtained, which integrates the semantic relevance of products at the content level with the co-occurrence frequency at the user behavior level. This reflects not only the inherent similarity of products in terms of function and attributes but also the potential connections between products in real-world purchasing scenarios, making the relationships between products more comprehensive and accurate, closely aligning with user needs and providing precise data support for subsequent personalized recommendations. The priority recommendation value of each candidate product is determined, and a recommendation result list is generated for display. This incorporates the user's historical purchasing preferences and uses the strong correlation between purchased and candidate products to personalize the matching coefficient. It considers both the relevance of search terms and products at the text level and incorporates the personalized needs reflected in the user's historical purchasing behavior. This allows the search results to be dynamically adjusted according to the user's consumption habits, naturally triggering intelligent recommendations while meeting current search needs, significantly improving user search efficiency, purchasing experience, and platform cross-selling conversion rates. Attached Figure Description

[0016] The following section provides a more detailed explanation of the procurement platform search optimization method based on intelligent recommendation algorithms proposed in this application, with reference to the accompanying drawings.

[0017] Figure 1 A flowchart illustrating the steps of a procurement platform search optimization method based on an intelligent recommendation algorithm provided in this application embodiment; Figure 2 A flowchart illustrating the steps of a method for obtaining synonym search terms provided in this application embodiment. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description, in conjunction with the accompanying drawings and implementation examples, provides a more comprehensive understanding of the procurement platform search optimization method and system based on intelligent recommendation algorithms proposed in this application. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0019] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains.

[0020] Please see Figure 1 The diagram illustrates a flowchart of a search optimization method for a procurement platform based on an intelligent recommendation algorithm, according to an embodiment of this application. The method includes the following steps: Step 1: In response to the current user's search terms, retrieve the platform's historical search logs, all users' historical purchase records, and product information for each item.

[0021] As the penetration rate of digital procurement among enterprises continues to increase, competition among procurement platforms has shifted from product richness to search accuracy and matching efficiency. By mining user behavior data and product semantic features, intelligent recommendation algorithms are introduced into procurement platforms to deeply explore the implicit synonyms of search terms, the multidimensional association features of products, and the personalized behavioral characteristics of users. This enables intelligent recall of search results and accurate personalized recommendations, thereby improving the accuracy of product retrieval and user experience on procurement platforms.

[0022] Based on the above analysis, by embedding tracking points in the front end of the procurement platform, the search terms of the current user can be captured; Analyze the historical search logs of the procurement platform's backend server to count the number of clicks and browsing duration for each product under each search term; and extract all users' historical purchase records from the server. By calling the API interface of the procurement platform, we can retrieve the product information for each product, including the product title text and the product details text; At this point, we have obtained the current user's search terms, all users' historical purchase records, the number of clicks and browsing duration of each product under each search term, as well as the product title text and product details text for each product.

[0023] Step 2: Based on the number of clicks and browsing time of products under any two search terms, as well as the overlap of clicked and viewed products, evaluate the consistency of user behavior between the two search terms, calculate the behavioral correlation between any two search terms, and thereby mine search terms that are synonymous with the current user's search terms, construct a search extension set, and filter out products whose product information contains search terms within the search extension set as candidate products.

[0024] When searching using user search terms, the same product may have different descriptions, meaning the search terms have synonyms. If text matching is performed solely based on search terms, a large number of relevant products will not be effectively retrieved. Therefore, by analyzing the click behavior of the same product under different search terms, the behavioral correlation between search terms is mined to extract synonyms for each search term, thereby achieving intelligent expansion of the search scope. Furthermore, the flowchart of the method for obtaining synonyms provided in this application embodiment is as follows: Figure 2 As shown, it specifically includes: Iterate through each search term in the historical search log, retrieve all products that have been clicked by the user under that search term, and form a set of clicked products. Find the intersection of the clicked product sets for any two search terms, and calculate the difference in the number of clicks for each product in the intersection under any two search terms, denoted as the click difference; In this embodiment, the absolute value of the difference in the number of clicks for each product within the intersection under any two search terms is calculated and denoted as the click difference.

[0025] Calculate the average browsing time of each product within the intersection under any two search terms, and record it as the average duration. The behavioral differences, frequency differences, and average duration of any two search terms are all positively correlated. It should be noted that a positive correlation means that the dependent variable increases as the independent variable increases and decreases as the independent variable decreases.

[0026] In this embodiment, the maximum value normalization method is used to normalize the frequency difference and the average duration. Specifically, the normalization process is as follows: the maximum number of clicks for each product under any two search terms is calculated, and the ratio of the frequency difference to the maximum value is taken as the normalized frequency difference; the maximum browsing time for each product under any two search terms is calculated, and the ratio of the average browsing time to the maximum browsing time is taken as the normalized average duration; the behavior difference is calculated as follows: the frequency differences of all products within the intersection are weighted and summed using the normalized average duration as the weight, and this sum is taken as the behavior difference.

[0027] It should be noted that the greater the difference in the number of clicks, the more significant the discrepancy in click preferences for the same product under the two search terms. The average duration reflects the user attention to the product under the two search terms; the higher the value, the stronger the product's appeal to users under both search terms. Thus, the greater the behavioral difference, the more significant the difference in click distribution for the more interested product under the two search terms, reflecting that the users' true needs and intentions behind the two search terms are not consistent.

[0028] Calculate the relevance between the clicked product sets of any two search terms, and use this as the product overlap. In this embodiment, the relevance is measured by calculating the Jaccard similarity of the clicked product sets of any two search terms. Jaccard similarity is a well-known technique and will not be described in detail here.

[0029] The behavioral relevance between any two search terms is positively correlated with the product overlap, but negatively correlated with the behavioral difference. It should be noted that a negative correlation means that the dependent variable decreases as the independent variable increases, and increases as the independent variable decreases.

[0030] In this embodiment, the ratio of product overlap to behavioral difference is normalized and used as the behavioral correlation degree. The maximum-minimum normalization method is used to normalize all ratios; this method is a well-known technique and will not be elaborated upon here. It should be noted that, to avoid a denominator of 0 when calculating the ratio, a parameter tuning factor is added to the denominator. In this embodiment, the parameter tuning factor is set to 1. In other implementation methods, the implementer can set the value according to the actual situation.

[0031] It should be noted that if the intersection is an empty set, the behavioral correlation degree is directly assigned to 0.

[0032] It should be noted that the greater the product overlap, the higher the overlap of products within the clicked product set under the two search terms, reflecting that the two search terms are more likely to be used to search for the same batch of products; the behavioral relevance reflects the overall similarity of the two search terms in terms of product coverage and user behavior. The larger the value, the more likely the two search terms are to point to the same user needs, that is, the more likely they are to be synonymous search terms.

[0033] For each search term, the remaining search terms that have a correlation with its behavior greater than or equal to a preset threshold are defined as synonym search terms; In this embodiment, the preset threshold value is 0.6. As for other implementation methods, the implementer can set it according to the actual situation.

[0034] Furthermore, by using search terms and their synonyms, the product information is matched to filter candidate products, specifically as follows: The current user's search terms and all their synonyms are combined to form a search extension set; each search term in the search extension set is traversed and matched with the product information of each product, and all products that are successfully matched are selected and defined as candidate products; In this embodiment, a string matching algorithm is used to match each search term in the search extension set with the product title text and product details text of each product. As long as the product title text or product details text of each product contains one of the search terms in the search extension set, the match is successful. The string matching algorithm is a well-known technology and will not be described in detail here.

[0035] At this point, all the items to be selected have been obtained.

[0036] Step 3: By searching the frequency and distribution of search terms within the extended set in the product information of the candidate products, determine the matching coefficient that represents the matching intent between each candidate product and the user's search intent.

[0037] Furthermore, by utilizing the distribution characteristics of search terms within the extended search set in the product information of the candidate products, the matching coefficient is calculated, specifically as follows: The total number of times all search terms within the extended search set appear in the product title text of each candidate product is used as the first match degree for each candidate product. The total number of times all search terms within the extended search set appear in the product details text of each candidate product is denoted as the term frequency factor. The product details text of each candidate product is divided into sentences. The number of sentences containing any search term in the search extension set is counted, and the proportion of each sentence to all sentences of the candidate product is calculated and recorded as the sentence coverage rate. In this embodiment, NLTK (Natural Language Toolkit) from the NLP toolkit is used for sentence segmentation. NLTK is a well-known technology and will not be described in detail here.

[0038] The product of sentence coverage and word frequency factor is used as the second matching degree for each candidate product; The first and second matching scores are positively fused to form the matching coefficient for each candidate product. In this embodiment, the specific process of forward fusion is as follows: The first matching degree and the second matching degree are normalized respectively. Using preset first weights and preset second weights as weights, the normalized first matching degree and the normalized second matching degree are weighted and summed to obtain the matching coefficient for each candidate product. The sum of the preset first weights and preset second weights is 1, and the preset first weight is greater than the preset second weight. In this embodiment, the preset first weight is 0.7 and the preset second weight is 0.3. In other implementation methods, the implementer can set these values ​​according to actual conditions. Next, the maximum value normalization method is used to normalize the results, obtaining the maximum first matching degree and the maximum second matching degree for all candidate products. The ratio of the first matching degree of each candidate product to the maximum first matching degree is used as the normalized first matching degree, and the ratio of the second matching degree of each candidate product to the maximum second matching degree is used as the normalized second matching degree.

[0039] It should be noted that the higher the first matching degree, the more frequently the current user's search terms and their synonyms appear in the product title of the candidate product, reflecting that the product is more relevant to the current user's search terms, the more the product matches the user's search intent, and the higher the probability that the user will pay attention to it; the higher the word frequency factor, the more frequently the search terms appear in the details text; the higher the sentence coverage, the more relevant the candidate product is to the overall content; the higher the second matching degree, the more fully and comprehensively the candidate product's details text describes the user's search terms, and the higher the matching coefficient, the higher the overall relevance of the candidate product to the user's search terms in terms of title and details description, reflecting that the candidate product better meets the user's needs in terms of text matching.

[0040] At this point, the matching coefficients for each candidate product are obtained.

[0041] Step 4: For all users' historical purchase records, based on the keyword relevance contained in the product information of any two products and the behavioral correlation when the keywords are retrieved, determine the semantic correlation of any two products at the content level. Combined with the frequency of the two products being purchased by users at the same time, obtain the correlation coefficient of any two products.

[0042] Furthermore, since users' historical purchasing behavior reflects their potential product preferences, text matching alone cannot fully reflect this personalized need. Therefore, by analyzing the co-purchase patterns of different products across all users' historical purchase records, and the correlation between product information across different products, a correlation coefficient is calculated, specifically: For all users’ historical purchase records, the percentage of users who purchased any two products at the same time is calculated as the product co-occurrence rate. In this embodiment, the specific calculation process of product co-occurrence is as follows: For any two products, assuming product A and product B, the sum of the total number of users who purchased product A and the total number of users who purchased product B is calculated, and the ratio of the number of users who purchased at the same time to the sum is used as the product co-occurrence.

[0043] It should be noted that the greater the co-occurrence of products, the higher the proportion of times the two products are purchased by the same user, reflecting that the two products are more closely related in the user's purchasing behavior and may be complementary or substitutes for each other.

[0044] For each product, the product information is segmented into words, all keywords are extracted, and the word vectors of each keyword are obtained. In this embodiment, after segmenting words and removing stop words using the Jieba word segmentation tool, keyword extraction is performed using the TF-IDF (Term Frequency-Inverse Document Frequency) method; word vectors are obtained using the Word2Vec model. Jieba word segmentation, the TF-IDF method, and the Word2Vec model are well-known technologies and will not be described in detail here.

[0045] For any two products in the historical purchase records of all users, let's say product A and product B; Choose any keyword from product A. Choose any keyword from product B. ,Will Define it as a keyword group; Calculate keywords in keyword groups With keywords The degree of correlation between word vectors is denoted as semantic similarity. In this embodiment, the relevance is calculated by keywords. With keywords The cosine similarity between word vectors is calculated and positively mapped to it as the semantic similarity. The positive mapping process is as follows: the sum of the cosine similarity and the value of 1 is used as the semantic similarity.

[0046] Get keywords and Behavioral relevance when used as a search term; It should be noted that, if using keywords or If no relevant record exists in the historical search logs of the procurement platform, the behavioral relevance of the keyword group to which the keyword belongs is directly assigned a value of 1. As another implementation method, the implementer can set it according to the actual situation.

[0047] The semantic similarity and behavioral relevance are positively integrated and used as the association contribution value of the keyword group; In this embodiment, the specific process of positive fusion is as follows: the product of semantic similarity and behavioral relevance is used as the association contribution value.

[0048] Keywords for product A Take the maximum correlation contribution value among all keywords in product B and the keyword group composed of it; The semantic correlation between any two products is positively correlated with the maximum correlation contribution value; In this embodiment, the average of the maximum association contribution values ​​of all keywords in product A is used as the semantic association degree between any two products. It should be noted that the greater the semantic similarity, the closer the two keywords are in the vector space, reflecting that the two keywords are semantically similar and may point to the same thing or the same attribute; the greater the behavioral correlation, the more the two keywords, when used as search terms, point to the same product and have a high degree of consistency in user behavior, reflecting that the two keywords can be substituted for each other in user search behavior and are more likely to be synonymous search terms; the greater the correlation contribution value, the more the keyword of product A has found the most matching keyword in product B, reflecting that the attribute or feature represented by the keyword has a good correspondence in product B; the greater the semantic correlation, the closer the two products are semantically at the keyword level as a whole, reflecting that the two products belong to similar fields and may meet similar user needs.

[0049] The correlation coefficient between any two products is positively correlated with the semantic correlation and the co-occurrence of the products; In this embodiment, the product of semantic relevance and product co-occurrence is used as the relevance coefficient between any two products.

[0050] It should be noted that the larger the correlation coefficient, the more likely the two products are to be purchased together and are highly similar in meaning, reflecting a very strong correlation between the two products, which is highly related at both the behavioral and content levels.

[0051] Thus, the correlation coefficient between any two products is obtained.

[0052] Step 5: Based on the correlation coefficients between different products and each candidate product in the current user's historical purchase records, and combined with the matching coefficients, determine the priority recommendation value of each candidate product, thereby generating a recommendation result list for intelligent recommendation and display.

[0053] Furthermore, by analyzing the matching coefficients of the candidate products selected under the current user's search terms and their correlation coefficients with products in the current user's historical purchase records, a priority recommendation value is determined, specifically as follows: For each item purchased in the current user's historical purchase record, obtain its correlation coefficient with each candidate item; and positively fuse the correlation coefficients between each candidate item and all items purchased by the current user as a personalization factor for each candidate item. In this embodiment, the sum of the correlation coefficients between each candidate product and all products currently purchased by the user is used as the personalization factor for each candidate product. The priority recommendation value of each candidate product is positively correlated with the matching coefficient and the personalization factor; In this embodiment, the formula for calculating the priority recommendation value of each candidate product is as follows: in, For the first The priority recommendation value of each candidate product. For the first The matching coefficient of each candidate product. For the first Personalization factors for each candidate product; It should be noted that the larger the personalization factor, the stronger the correlation between the candidate product and the user's historical purchases, the more the product matches the user's historical preferences, and the more worthy of recommendation. The larger the priority recommendation value, the more the candidate product matches the user's search terms at the text level and is closely related to the user's historical purchases at the user preference level, and should be ranked higher to be displayed to the user first.

[0054] Based on the current user's search terms, all candidate products are sorted in descending order according to their priority recommendation value to generate a list of recommended results. This list is then displayed on the current user's search interface for browsing.

[0055] Based on the same inventive concept as the above methods, this application also provides a procurement platform search optimization system based on intelligent recommendation algorithms, including a memory, a processor, and a computer program stored in the memory and running on the processor. When the processor executes the computer program, it implements the steps of any one of the above-described procurement platform search optimization methods based on intelligent recommendation algorithms.

[0056] It should be understood that, although Figure 1 The steps in the flowchart are shown sequentially as indicated by the arrows, but these steps are not necessarily executed in the order indicated by the arrows. Unless otherwise specified herein, there is no strict order in which these steps are executed, and they can be performed in other orders. Figure 1At least some of the steps in the process may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these sub-steps or stages is not necessarily sequential, but can be executed in turn or alternately with other steps or at least some of the sub-steps or stages of other steps.

[0057] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0058] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application. Therefore, any simple modifications, equivalent changes, and alterations made to the above embodiments based on the technical essence of this application, without departing from the content of the technical solution of this application, shall fall within the protection scope of the technical solution of this application.

Claims

1. A search optimization method for a procurement platform based on an intelligent recommendation algorithm, characterized in that, The method includes the following steps: Extract product information and historical search logs recorded on the platform. Based on the number of clicks and browsing time of products under any two search terms in the logs, as well as the overlap of clicked and browsed products, evaluate the consistency of user behavior between the two search terms, calculate the behavioral correlation between any two search terms, and thereby mine search terms that are synonymous with the current user's search terms. Construct a search extension set, and filter out products whose product information contains search terms within the search extension set as candidate products. By retrieving the frequency and distribution of search terms within the extended set in the product information of the candidate products, the matching coefficient representing the user's search intent for each candidate product is determined. Based on the historical purchase records of all users, the semantic correlation of any two products at the content level is determined according to the keyword relevance contained in the product information between any two products and the behavioral correlation when the keywords are searched. The correlation coefficient of any two products is obtained by combining the frequency of the two products being purchased by users at the same time. Based on the correlation coefficients between different products and each candidate product in the current user's historical purchase records, combined with the matching coefficients, the priority recommendation value of each candidate product is determined, and a recommendation result list is generated for intelligent recommendation and display.

2. The procurement platform search optimization method based on intelligent recommendation algorithm as described in claim 1, characterized in that, The calculation of the behavioral relevance between any two search terms includes: Traverse each search term in the historical search log, obtain all products clicked by the user under the search term, and form a set of clicked products; find the intersection of the clicked product sets of any two search terms, analyze the differences in click and browsing behavior of the same products under the two search terms within the intersection, and determine the degree of behavioral difference between the two search terms. Calculate the relevance between the clicked product sets of any two search terms, and use this as the product overlap. The behavioral correlation is positively correlated with the product overlap, but negatively correlated with the behavioral difference.

3. The procurement platform search optimization method based on intelligent recommendation algorithm as described in claim 2, characterized in that, The calculation process for the behavioral difference is as follows: calculate the difference in the number of clicks for each product within the intersection under any two search terms, and record it as the number difference; calculate the average browsing time for each product within the intersection under any two search terms, and record it as the average duration; the behavioral difference is positively correlated with both the number difference and the average duration.

4. The procurement platform search optimization method based on intelligent recommendation algorithm as described in claim 1, characterized in that, The process of constructing the search extension set is as follows: for each search term, the remaining search terms whose relevance to its behavior is greater than or equal to a preset threshold are defined as synonyms, and the current user's search terms and all its synonyms are combined to form the search extension set.

5. The procurement platform search optimization method based on intelligent recommendation algorithm as described in claim 1, characterized in that, The matching coefficient is calculated as follows: The product information includes product title text and product details text; the total number of times all search terms in the search extension set appear in the product title text of each candidate product is counted as the first match degree of each candidate product; The total number of times all search terms in the extended search set appear in the product details text of each candidate product is recorded as the term frequency factor; the product details text of each candidate product is divided into sentences, and the number of sentences containing any search term in the extended search set is counted, and the proportion of each sentence to all sentences of the candidate product is calculated, which is recorded as the sentence coverage rate; The product of sentence coverage and word frequency factor is used as the second matching degree for each candidate product; The matching coefficient is the result of a positive fusion of the first matching degree and the second matching degree.

6. The procurement platform search optimization method based on intelligent recommendation algorithm as described in claim 1, characterized in that, The process for determining the semantic relevance is as follows: For each product, the product information is segmented into words, all keywords are extracted, and the word vectors of each keyword are obtained. For all users' historical purchase records, select one keyword from each of the two products and define them as a keyword group; calculate the correlation between the word vectors of the two keywords in the keyword group, and record it as semantic similarity; obtain the behavioral relevance when using the two keywords in the keyword group as search terms; positively fuse semantic similarity and behavioral relevance as the relevance contribution value of the keyword group; For any two products, for any keyword of one product, select the maximum association contribution value of the keyword group formed by that keyword and all keywords of the other product; The semantic correlation between any two products is positively correlated with the maximum correlation contribution value.

7. The procurement platform search optimization method based on intelligent recommendation algorithm as described in claim 1, characterized in that, The method of obtaining the correlation coefficient between any two products includes: calculating the percentage of users who purchased any two products simultaneously based on the historical purchase records of all users, which is used as the product co-occurrence degree; the correlation coefficient is positively correlated with both semantic correlation degree and product co-occurrence degree.

8. The procurement platform search optimization method based on intelligent recommendation algorithm as described in claim 1, characterized in that, The determination of the priority recommendation value for each candidate product includes: For each product purchased in the current user's historical purchase record, obtain its correlation coefficient with each candidate product; and positively fuse the correlation coefficients between each candidate product and all products purchased by the current user as a personalization factor for each candidate product. The priority recommendation value is positively correlated with the matching coefficient and the personalization factor.

9. The procurement platform search optimization method based on intelligent recommendation algorithm as described in claim 1, characterized in that, The process of obtaining the recommended results list is as follows: for the current user's search terms, all candidate products are sorted in descending order according to their priority recommendation value to generate a recommended results list.

10. A procurement platform search optimization system based on an intelligent recommendation algorithm, comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the procurement platform search optimization method based on intelligent recommendation algorithm as described in any one of claims 1-9.

Images