A commodity information automatic completion method based on semantic association

Abstract

The application relates to the technical field of information analysis, and discloses a commodity information automatic completion method based on semantic association, which comprises the following steps: obtaining all commodity information of a target commodity; judging whether the commodity information of the target commodity is missing; if the commodity information is missing, completing the commodity information; and executing a dynamic adjustment commodity information display strategy. The commodity information automatic completion method based on semantic association scientifically judges whether the commodity information is missing by comprehensively analyzing user behavior data and purchase conversion data, avoids subjective speculation, ensures the effectiveness of the completion measures, accurately locates the specific part of the information missing, provides a clear direction for the subsequent completion measures, avoids blind completion, meets the demand of users for commodity information by supplementing and improving the commodity information, reduces the confusion and hesitation of the users when browsing the commodity page, improves the user satisfaction, optimizes the commodity information display, and improves the purchase conversion rate.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of information analysis, in particular to a commodity information automatic completion method based on semantic association. BACKGROUND

[0002] With the rapid development of e-commerce, commodity information, as a key factor in consumer decision-making, its integrity and accuracy become crucial. Traditional commodity information entry and maintenance methods often rely on manual completion, which has problems such as low efficiency, easy errors, inconsistent information, high maintenance costs, etc. In order to solve these problems, researchers and enterprises have begun to explore commodity information automatic completion technology. This technology uses artificial intelligence and natural language processing technology to automatically obtain information from various sources and complete missing commodity information, thereby improving information quality and efficiency. The commodity information automatic completion method based on semantic association integrates advanced natural language processing technology, knowledge graph technology, and semantic association analysis algorithm to realize more intelligent, accurate, and interpretable commodity information automatic completion, thereby improving the quality of commodity information, enhancing user experience, and providing stronger data support for e-commerce platforms.

[0003] The existing commodity information automatic completion method based on semantic association cannot comprehensively analyze user behavior data and purchase conversion data to scientifically determine whether the commodity information is missing, which is prone to subjective speculation, resulting in ineffective information completion, inability to accurately locate the specific part of the missing information, and inability to provide a clear direction for subsequent completion measures. It is prone to blind completion and cannot meet the needs of users for commodity information, increasing users' confusion and hesitation when browsing commodity pages, reducing user satisfaction, and thus reducing purchase conversion rates. Its practicality has certain limitations. SUMMARY

[0004] The present application provides a commodity information automatic completion method based on semantic association to promote the solution to the problems in the background art.

[0005] The present application provides the following technical solution: a commodity information automatic completion method based on semantic association, comprising:

[0006] Obtain all commodity information of the target commodity, and generate a commodity information partial set: ;

[0007] Determine whether the commodity information of the target commodity is missing;

[0008] If the commodity information of the target commodity is not missing, do not perform the information completion operation;

[0009] If the commodity information of the target commodity is missing, complete the commodity information of the target commodity, and perform a dynamic adjustment of the commodity information display strategy.

[0010] The execution dynamic adjustment commodity information display strategy, specifically:

[0011] For each part of the target commodity information , calculate its interest degree weight:

[0012] ;

[0013] For each part of the target commodity information , calculate its confusion degree weight:

[0014] ;

[0015] For each part of the target commodity information , according to its interest degree weight and confusion degree weight, determine the display content:

[0016] .

[0017] As an optional solution of the application, the method for automatically completing the commodity information based on semantic association, wherein: judging whether the target commodity information is missing, including analyzing all users browsing the target commodity in turn, specifically:

[0018] Obtain the database ;

[0019] Obtain the identification of the target commodity, called target identification, denoted as ;

[0020] Extract all records of the target identification in the database, denoted as ;

[0021] Define a record analysis function to judge whether the target commodity has browsing records:

[0022] ;

[0023] If , it is determined that the target commodity has browsing records;

[0024] Then query the user behavior database corresponding to the target identification in the database, get all the records corresponding to the target commodity in the user behavior database, form a browsing record set:

[0025] ;

[0026] Generate a browsing user set: ;

[0027] If If so, it is determined that the target product has no browsing history;

[0028] Then, query the user behavior logs corresponding to the target identifier in the database and generate a set of query results:

[0029] ;

[0030] Set up a query analysis function to determine if a collection of browsing users can be generated: ;like If so, it is determined that a set of browsing users can be generated; if If so, it is determined that a browsing user set cannot be generated;

[0031] Generate a collection of browsing users: .

[0032] As an optional solution to the semantic association-based automatic product information completion method of the present invention, the method further includes: sequentially analyzing all users who browse the target product, and further comprising:

[0033] Get the collection of browsing users ;

[0034] For the browsing user set Each user in Extract detailed browsing behavior data when browsing the target product page;

[0035] Define an attention function Calculate users For each product information section The level of attention paid to the product information during the browsing process is analyzed to identify the areas of focus for each user.

[0036] ;

[0037] Define a perplexity function Calculate users For each product information section The level of confusion is used to identify potential points of confusion for users regarding different parts of the product information during the browsing process.

[0038] ;

[0039] Set up an attention analysis function to determine the user. Key considerations when browsing a target product page:

[0040] ;

[0041] like Then determine the user Product information section has higher attention;

[0042] If , it is determined that the user has lower attention;

[0043] A confusion degree analysis function is set to determine the confusion points that the user may have when browsing the target product page:

[0044] If , it is determined that the user may have confusion in the product information part

[0045] If , it is determined that the user has no confusion in the product information part

[0046] According to the results of the attention degree analysis function of each product information part , a set of attention points of the user is generated, denoted as :

[0047] According to the results of the confusion degree analysis function of each product information part , a set of confusion points of the user is generated, denoted as :

[0048] The set of attention points and the set of confusion points are integrated to generate the analysis result of the browsing behavior of the user .

[0049] As an optional solution of the product information automatic completion method based on semantic association, the method comprises the following steps:

[0050] A set of browsing users is obtained

[0051] Users in the set of browsing users who add the target product to the shopping cart are extracted to generate a set of added users, denoted as

[0052] The added rate of the target product is calculated, denoted as :​​​​​​​​ ;

[0053] Extract the users in the add-to-cart user set who enter the settlement page to generate a settlement user set, denoted as ;

[0054] Calculate the settlement rate of the target product, denoted as : ;

[0055] Extract the users in the settlement user set who successfully complete payment to generate a payment user set, denoted as ;

[0056] Calculate the payment rate of the target product, denoted as : ;

[0057] Calculate the purchase conversion rate of the target product, denoted as : ;

[0058] Count the number of users in the browsing user set who have the confusion point set , and calculate the confusion user proportion : ;

[0059] Count the frequency of the confusion point in the key information part of the user :

[0060] ;

[0061] Get the best-selling product set of the same category, denoted as : ;

[0062] Calculate the difference between the target product and the best-selling product in the same category in each part of the product information :

[0063] ;

[0064] Define a comprehensive analysis function to comprehensively judge whether the product information of the target product is missing:

[0065] ;

[0066] If , it is determined that the product information of the target product is missing;

[0067] If , it is determined that the product information of the target product is not missing.

[0068] As an optional solution to the semantic association-based automatic product information completion method of the present invention, the completion of product information for the target product includes analyzing the user profile of the target product, specifically:

[0069] Get the browsing user set ;

[0070] Extract browsing user set Each user in User profile data:

[0071] ;

[0072] Based on the browsing user set Each user in User profile data is used to segment users:

[0073] ;

[0074] For each user group, calculate their attention level to each part of the product information and analyze their engagement with the product information. Focus and needs: ;

[0075] Based on the attention levels of each user group, supplementary product information is generated for each part of the product information, denoted as... :

[0076] .

[0077] As an optional solution to the semantic association-based automatic product information completion method of the present invention, the method further includes: completing the product information of the target product, and constructing and analyzing a user behavior model, specifically:

[0078] Get the browsing user set ;

[0079] Extract browsing user set Each user in Behavioral data while browsing the target product page;

[0080] For each user Define its behavioral feature vector : ;

[0081] For each user Construct a behavior path matrix : ;

[0082] For each product information section , calculate its frequency of being accessed : ;

[0083] Find the high-frequency key path of the user in the browsing behavior of the target commodity :

[0084] ;

[0085] For each commodity information part , calculate its information demand degree , consider the node access frequency and the behavior characteristics on the key path: ;

[0086] Define a gap identification function to determine whether the commodity information part has information gaps:

[0087] ;

[0088] If , it is determined that the commodity information part has information gaps;

[0089] If , it is determined that the commodity information part does not have information gaps;

[0090] According to the analysis results of each part of the commodity information under the gap identification function, generate the supplemented commodity information for each part of the commodity information, denoted as :

[0091] .

[0092] As an optional solution of the commodity information automatic completion method based on semantic association, wherein: the commodity information of the target commodity is supplemented, further comprising mining user behavior association rules, specifically:

[0093] Obtain a set of browsing users ;

[0094] Extract the behavior data of each user when browsing the target commodity page from the set of browsing users ;

[0095] For each user , define its behavior sequence: ;

[0096] The browsing behavior of each user is regarded as a transaction, and a behavior transaction database is constructed:

[0097] ;

[0098] Mining frequent itemsets from transaction databases ; ;

[0099] Generating association rules from frequent itemsets in the form of ;

[0100] Defining the confidence of association rules : ;

[0101] For association rules , define a gap checking function to check whether there is missing information related to frequent subsequent behavior in the product information: ;

[0102] If , it is determined that there is information gap in this part of the product information;

[0103] If , it is determined that there is no information gap in this part of the product information;

[0104] According to the association rules and the analysis results of each part of the product information under the gap checking function, generate the supplemented product information for each part of the product information, denoted as :

[0105] .

[0106] As an optional solution of the semantic association-based automatic product information completion method, wherein: the product information of the target product is completed, further comprising cross-platform data integration and analysis, specifically:

[0107] Set up a multi-channel data source set: ;

[0108] Integrate the multi-channel data source set and all user behavior data and feedback information to generate an integrated data set: ;

[0109] Get a set of browsing users ;

[0110] Extract the behavior data of each user in the set of browsing users when browsing the target product on different platforms;

[0111] For each user , define its behavior vector on different platforms: ;

[0112] Identify a set of browsing users Hot topics of users in the middle;

[0113] Extract a set of user comments ;

[0114] Extract a set of hot topics ;

[0115] For each user comment , analyze its sentiment tendency :

[0116] ;

[0117] Define a missing analysis function to determine the missing part of the product information of the target product:

[0118] ;

[0119] If , it is determined that the product information of the target product is missing Part;

[0120] If , it is determined that the product information of the target product is not missing Part;

[0121] According to the analysis result of each part of the product information under the missing analysis function, the supplemented product information of each part of the product information is generated, denoted as :

[0122] .

[0123] The present application has the following advantages:

[0124] 1. The product information automatic completion method based on semantic association, by querying the user behavior database or user behavior log, all users who have browsed the target product are obtained, and the browsing behavior data is analyzed for each user who has browsed the target product in turn, covering all users who have browsed the target product, avoiding missing any potential user feedback, accurately identifying the focus and confusion points of users in the browsing process, analyzing the behavior of each user, and being able to find the differentiated needs of different user groups.

[0125] 2、The method automatically completes the missing information of the target product based on semantic association. It collects data of each link in the purchase conversion process of the target product, including the rate of adding to the shopping cart, the rate of entering the settlement page, the rate of payment success, and the purchase conversion rate, etc. It determines the purchase conversion data of the target product, comprehensively analyzes the user browsing behavior data and the purchase conversion data, and judges whether the product information of the target product is missing. It fully evaluates the user's behavior in the purchase process, finds the potential conversion bottleneck, ensures that the completion measures can effectively improve the purchase conversion rate, scientifically judges whether the product information is missing, avoids subjective speculation, accurately locates the specific part of the missing information, and provides a clear direction for subsequent completion measures.

[0126] 3、The method automatically completes the missing information of the target product based on semantic association. It collects user portrait data of the target product, including basic information such as age, gender, region, consumption level, purchase preference, browsing history, and evaluation history. According to the user portrait data, it analyzes the attention points and demand differences of different user groups to the product information, supplements and perfects the product information according to the attention points and demands of different user groups, meets the user's demand for product information, improves the user's experience when browsing the product page, reduces the user's confusion and hesitation in the purchase process, thereby improving the purchase conversion rate, continuously optimizing the display of product information, and ensuring that the information display effect is always in the best state.

[0127] 4、The method automatically completes the missing information of the target product based on semantic association. Based on the browsing behavior data of the browsing user set, it constructs a user behavior model to describe the behavior characteristics and path of the user when browsing the target product page. It analyzes the key nodes and behavior patterns in the user behavior model, finds out the information demand points and possible information gaps in the browsing process, and supplements the missing parts of the product information according to the analysis results of the user behavior model, meets the user's demand for product information, improves the user's experience when browsing the product page, reduces the user's confusion and hesitation in the purchase process, thereby improving the purchase conversion rate, continuously optimizing the display of product information, and ensuring that the information display effect is always in the best state.

[0128] 5、The method automatically completes the missing information of the target product based on semantic association. It uses data mining technology such as association rule mining algorithm to mine the association relationship between user behavior data, supplements the missing information related to key behaviors in the product information according to the mined association rules, meets the user's demand for product information, improves the user's experience when browsing the product page, reduces the user's confusion and hesitation in the purchase process, thereby improving the purchase conversion rate, continuously optimizing the display of product information, and ensuring that the information display effect is always in the best state.

[0129] 6、The method for automatically completing product information based on semantic association integrates user behavior data and feedback information from multiple channels such as e-commerce platforms, social media platforms, and offline physical stores, analyzes the discussion hotspots, focus points, and question points of users on the target product in cross-platform data, supplements the missing parts of product information based on the results of cross-platform data integration and analysis, meets the user's demand for product information, improves the user's experience when browsing the product page, reduces the user's confusion and hesitation in the purchase process, thereby improving the purchase conversion rate, continuously optimizing product information display, and ensuring that the information display effect is always in the best state.

[0130] 7、The method for automatically completing product information based on semantic association dynamically adjusts the display mode and content of product information based on user behavior analysis results, continuously optimizes product information display, ensures that the information display effect is always in the best state, meets the user's demand for product information by supplementing and perfecting product information, improves the user's experience when browsing the product page, reduces the user's confusion and hesitation in the purchase process by optimizing product information display, thereby improving the purchase conversion rate. BRIEF DESCRIPTION OF DRAWINGS

[0131] Figure 1 The flowchart of the method for automatically completing product information based on semantic association. DETAILED DESCRIPTION

[0132] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, not all. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0133] Embodiment one, a method for automatically completing product information based on semantic association, referring to Figure 1 , comprising:

[0134] Obtain all product information of the target product, generate a product information partial set, the product information is information used to describe the specific circumstances of the product, including the picture, material, structure, etc. of the product:

[0135] ; wherein, represents the th part of the product information, ;

[0136] Determine whether the product information of the target product is missing;

[0137] If the product information of the target product is not missing, do not perform the information completion operation;

[0138] If the product information of the target product is incomplete, the product information of the target product is completed, and a dynamic adjustment of the product information display strategy is performed;

[0139] The dynamic adjustment of the product information display strategy is specifically:

[0140] For each part of the product information of the target product , its interest degree weight is calculated:

[0141] ;

[0142] Wherein, represents the user 's stay time in the product information part , represents the user 's click times in the product information part , represents the user 's scroll depth in the product information part , , , are weight coefficients of stay time, click times and scroll depth respectively, and , used to balance the influence of behavior characteristics such as stay time, click times and scroll depth on interest degree weight, by adjusting these weights, the importance of certain behavior characteristics can be emphasized, so that they occupy a larger proportion in the calculation of interest degree weight, represents the user 's stay time in the product information part , represents the user 's click times in the product information part , represents the user 's scroll depth in the product information part , , , are weight coefficients of stay time, click times and scroll depth respectively, and , used to balance the influence of behavior characteristics such as stay time, click times and scroll depth on interest degree weight, by adjusting these weights, the importance of certain behavior characteristics can be emphasized, so that they occupy a larger proportion in the calculation of interest degree weight, represents the specific product information part that is currently being analyzed or considered for adjustment of the display strategy, which is the focus part for attention when calculating the interest degree weight, used to judge whether this part needs to be adjusted in the display mode according to the user behavior data, represents all other parts in the commodity information, used as a reference for comparison or normalization in the calculation of the interest degree weight, that is, as a reference system to participate in the calculation, ensuring that the calculated interest degree weight has relative significance and comparability;

[0143] For each part of the commodity information of the target commodity , calculate its confusion degree weight:

[0144] ;

[0145] Wherein, represents the number of times the user clicks the back button in the commodity information part , represents the number of times the user jumps the browsing path in the commodity information part , , are the weights of the number of times the back button is clicked and the number of times the browsing path is jumped, respectively, and , used to balance the influence of the number of times the back button is clicked and the number of times the browsing path is jumped on the confusion degree, the contribution of different confusion behaviors in the confusion degree calculation can be adjusted according to actual needs, represents the number of times the user clicks the back button in the commodity information part , represents the number of times the user jumps the browsing path in the commodity information part , , are the weights of the number of times the back button is clicked and the number of times the browsing path is jumped, respectively, and , used to balance the influence of the number of times the back button is clicked and the number of times the browsing path is jumped on the confusion degree, the contribution of different confusion behaviors in the confusion degree calculation can be adjusted according to actual needs;

[0146] For each part of the commodity information of the target commodity , determine the display content according to its interest degree weight and confusion degree weight:

[0147] ;

[0148] Wherein, represents highlighting the commodity information part , represents simplified display of the commodity information part , represents displaying the original content of the commodity information part , is a threshold of the interest weight, used to determine whether the product information part needs to be highlighted. When the interest weight of the product information part, i.e., the relative value of the degree of attention of the user to the part relative to all product information parts, is greater than or equal to the threshold, it is considered that the part has a high attraction to the user, and should be highlighted to further attract the attention of the user and enhance the attention of the user to the part, is a threshold of the confusion weight, used to determine whether the product information part needs to be simplified or optimized. When the confusion weight of the product information part, i.e., the relative value of the degree of confusion of the user when browsing the part relative to all product information parts, is greater than or equal to the threshold, it is considered that the part may have problems, causing the user to be confused, and at this time the display content of the part should be considered to be simplified or optimized to improve the user experience and reduce the confusion of the user when browsing the part.

[0149] Wherein, judging whether the product information of the target product is missing includes sequentially analyzing all users browsing the target product, specifically:

[0150] Obtaining a database , the database is a database storing all products on the shopping platform, all product information of each product, all users of the platform, and browsing and consumption of each user;

[0151] Obtaining the identification of the target product, denoted as target identification, denoted as ;

[0152] Extracting all records of the target identification in the database, denoted as ;

[0153] Defining a record analysis function to determine whether the target product has browsing records:

[0154] ;

[0155] Wherein, represents an empty set, i.e., the target identification has no records in the database, indicating that the target product may have no browsing records, or the browsing records of the target product are recorded in the user behavior log;

[0156] If , it is determined that the target product has browsing records;

[0157] Then query the user behavior database corresponding to the target identification in the database to obtain all records corresponding to the target product in the user behavior database, forming a browsing record set:

[0158] ;

[0159] Wherein, is a query function used to query the corresponding user behavior database of the target identifier in the database, is a query function used to query the corresponding user behavior database of the target identifier in the database;

[0160] Generate a browsing user set: ;

[0161] wherein, represents the browsing behavior data of the user to the target commodity , if the user has browsed the target commodity , the value is greater than or equal to 1, otherwise 0;

[0162] If , it is determined that the target commodity does not exist browsing record;

[0163] Then query the corresponding user behavior log of the target identifier in the database to generate a query result set:

[0164] ;

[0165] wherein, is the user behavior log corresponding to the target identifier, is a query function used to query the corresponding user behavior log of the target identifier in the database;

[0166] Set a query analysis function to determine whether the browsing user set can be generated:

[0167] ;

[0168] If , it is determined that the browsing user set can be generated;

[0169] If , it is determined that the browsing user set cannot be generated;

[0170] Generate a browsing user set:

[0171] ;

[0172] wherein, represents the browsing behavior data of the user to the target commodity , if the user has browsed the target commodity , the value is greater than or equal to 1, otherwise 0, represents an empty set, that is, the target identifier does not exist browsing record at the same time in the user behavior log, which means that the target commodity has no browsing record.

[0173] This includes analyzing all users who browsed the target product sequentially, and also includes:

[0174] Get the collection of browsing users ;

[0175] For the browsing user set Each user in Extract detailed browsing behavior data when browsing the target product page, including but not limited to dwell time, browsing path, mouse movement trajectory, click location, number of clicks, scroll depth, number of images viewed, and number of times the details page is expanded;

[0176] Define an attention function Calculate users For each product information section The level of attention paid to the product information during the browsing process is analyzed to identify the areas of focus for each user.

[0177] ;

[0178] in, Indicates user In the product information section The length of stay Indicates user In the product information section Number of clicks Indicates user In the product information section The scroll depth, which is the percentage of the page that has scrolled. , , The weights are respectively for dwell time, number of clicks, and scroll depth, and This is used to balance the impact of behavioral features such as dwell time, click count, and scroll depth on interest score weights. By adjusting these weights, the importance of certain behavioral features can be emphasized. Represents a collection of browsing users The first in One user, The first part representing product information Each part;

[0179] Define a perplexity function Calculate users For each product information section The level of confusion is used to identify potential points of confusion for users regarding different parts of the product information during the browsing process.

[0180] ;

[0181] in, representing the user the number of times of clicking the back button in the commodity information part , representing the user the number of times of jumping the browsing path in the commodity information part , , respectively the weight of the number of times of clicking the back button and the number of times of jumping the browsing path, and , used to balance the influence of the number of times of clicking the back button and the number of times of jumping the browsing path on the confusion degree, the contribution of different confusion behaviors in identifying the confusion point can be adjusted according to actual needs;

[0182] setting an attention analysis function to determine the attention point of the user when browsing the target commodity page:

[0183] ;

[0184] wherein, is the attention threshold, used to judge whether the attention of the user to a certain commodity information part is high enough, so as to decide whether the part needs to be highlighted or preferentially displayed;

[0185] if , it is determined that the user has a higher attention to the commodity information part ;

[0186] if , it is determined that the user has a lower attention to the commodity information part ;

[0187] setting a confusion analysis function to determine the possible confusion point of the user when browsing the target commodity page: ;

[0188] wherein, is the confusion threshold, used to judge whether the commodity information part needs to be simplified or optimized, when the confusion degree is greater than or equal to the confusion threshold, it indicates that the user encounters great confusion when browsing the commodity information part, which may be caused by unclear, incomplete or too complex information, etc., leading to the user's difficulty in understanding or finding the required content, at this time, the content needs to be simplified or optimized to reduce the user's confusion and improve the user experience, when the confusion degree is less than the confusion threshold, it means that the confusion degree of the user when browsing the part is within an acceptable range, and no special adjustment is needed, the original display mode can be maintained;

[0189] if , it is determined that the user In the commodity information part There may be confusion;

[0190] If , it is determined that the user In the commodity information part There is no confusion;

[0191] According to the result of the attention analysis function of each commodity information part , the user Generate a set of attention points, denoted as : ;

[0192] According to the result of the confusion analysis function of each commodity information part , the user Generate a set of confusion points, denoted as : ;

[0193] Integrate the attention point set and the confusion point set , generate the user The analysis result of the browsing behavior.

[0194] In this embodiment, it is determined whether the commodity information of the target commodity is missing, which includes comprehensive analysis of user browsing behavior data and purchase conversion whole process data to determine whether the commodity information of the target commodity is missing, specifically:

[0195] Get the set of browsing users ;

[0196] Extract the users in the set of browsing users Add the target commodity to the shopping cart, generate the set of added users, denoted as ;

[0197] Calculate the added rate of the target commodity, denoted as : ; wherein The number of elements in the set of added users, The number of elements in the set of browsing users ;

[0198] Extract the users in the set of added users who enter the settlement page, generate the set of settlement users, denoted as ;

[0199] Calculate the settlement rate of the target commodity, denoted as : ; wherein The number of elements in the set of settlement users;

[0200] Extracting users who successfully complete payment in the settlement user set, generating a payment user set, denoted as ;

[0201] Calculating the payment rate of the target product, denoted as : ; wherein, is the number of elements in the payment user set;

[0202] Calculating the purchase conversion rate of the target product, denoted as : ;

[0203] Counting the number of users in the browsing user set who have a confusion point set , calculating the confusion user proportion : ; wherein, is an indicator function, taking the value 1 if the user has a confusion point set , i.e., the user has a confusion point set , and taking the value 0 otherwise, is the number of browses;

[0204] Counting the confusion point occurrence frequency of users in the key information part , the key information part refers to the part of the product information that has a significant impact on user purchase decision and use experience. These parts are usually the most concerned content of users when browsing the product page, and may also be the key factors affecting whether the user purchases the product, including product description, product parameters, user evaluation, application scenarios, after-sales service, price information, brand information, etc.

[0205] ; wherein, is the number of key parts of the target product information, the number of key parts refers to the number of key information parts in the target product information, i.e., the number of elements in the key information part set , which reflects the number of parts that need to be focused on and optimized in the product information, represents is a key information part, is an indicator function, taking the value 1 if the user 's confusion point set contains the key information part , i.e., the confusion point set and the key information part overlap, and taking the value 0 otherwise;

[0206] Obtain a collection of similar best-selling products, denoted as : ;

[0207] Calculate the differences between the target product and similar best-selling products in various parts of the product information. :

[0208] ;

[0209] in, This indicates that the target product is in the product information section. Part of the content, Indicates best-selling products in the same category In the product information section Part of the content, This is an indicator function; if the target product is similar to best-selling products of the same type... In the product information section If some parts of the content are different, the value is 1; otherwise, the value is 0. To determine the quantity of similar best-selling products, when calculating the differences between the target product and similar best-selling products in various parts of product information, we focus on the substantial differences in content, rather than literal identicalness. We can determine whether there are substantial differences in content based on semantic similarity or information structure. Determining substantial differences based on semantic similarity involves using text similarity calculation methods (such as cosine similarity, Jaccard similarity, etc.) to measure whether the content of two products is similar in a certain information section. If the similarity is below a certain threshold, the content is considered different. Determining substantial differences based on information structure involves comparing whether the information sections of two products contain the same fields or structured information. If any fields are missing or the structure is different, the content is considered different. For example, suppose there is a target product and similar best-selling products... We need to compare them in the product information. The product information section includes a description of features, user reviews, and applicable scenarios. The target product's feature description states "possesses high-speed processing capabilities and supports multitasking." Similar best-selling products... The function description states "supports fast multitasking," and the two descriptions have high semantic similarity. Therefore, they are determined to be identical, and 1 is output. The target product's user reviews include a review summary and detailed reviews. Similar best-selling products... The user reviews only contain a summary of the reviews, and since their information structures are different, they are determined to be different, so the output is 0. The content of the target product in the applicable scenarios lists multiple usage scenarios and similar best-selling products. The applicable scenarios are not listed in detail, and the compared samples have missing content. Therefore, the two are determined to be different, and the output is 0.

[0210] Define a comprehensive analysis function to comprehensively judge whether the product information of the target product is missing:

[0211] ;

[0212] Wherein, is the proportion of confused users threshold, which is used to measure whether the proportion of users who are confused about the product information among the users who browse the target product page is too high. If the actual proportion of confused users exceeds this threshold, it indicates that the product information may have problems and needs to be further analyzed whether there is information missing. For example, if the proportion of confused users threshold is 30%, is the purchase conversion rate threshold, which is used to determine whether the proportion of the final purchase users of the target product among the users who browse the product page is too low. A lower purchase conversion rate may indicate that the product information is not enough to attract users to purchase, and there may be missing or unclear key information, which affects the user's purchase decision. For example, if the purchase conversion rate threshold is 10%, is the frequency of confusion point threshold, which is used to evaluate whether the frequency of confusion points when users browse the key information part is too high. When the frequency of confusion points exceeds this threshold, it means that the key part of the product information may have unclear or missing expressions, which leads to user understanding difficulties and needs to be optimized, is the difference threshold, which is used to compare the difference degree of the target product and the best-selling product in the product information. If the difference degree exceeds the set threshold, it may mean that the target product has deficiencies in information display, and needs to refer to the practice of the best-selling product to complete or optimize the information.

[0213] If , it is determined that the product information of the target product is missing.

[0214] If , it is determined that the product information of the target product is not missing.

[0215] Wherein, the collection process of the key information part is as follows:

[0216] For each product information part , calculate the ratio of the number of users accessing this part in the set of browsing users to the total number of users: ;

[0217] For each product information part , calculate the average time spent in this part by the set of browsing users : ;

[0218] For each product information part , count the set of browsing users The number of clicks and the number of interaction behaviors in this section:

[0219] ;

[0220] Where, The number of clicks in this product information section, The number of interaction behaviors in this product information section;

[0221] Calculate the TF-IDF value of each keyword using the TF-IDF algorithm:

[0222] ;

[0223] Where, ;

[0224] ;

[0225] Extract high-frequency keywords from user comments and discussions:

[0226] ;

[0227] Where, TF-IDF filtering threshold, used to extract high-frequency keywords from all keywords;

[0228] Use LDA and other topic modeling algorithms to identify the main topics in user comments and discussions:

[0229] ;

[0230] Analyze user behavior in the purchase conversion funnel and calculate the conversion rate at each stage:

[0231] ;

[0232] ​​​​​The purchase conversion funnel is a model used to describe the entire process from a user's initial contact with a product to the final purchase. It divides the user's purchase behavior into multiple stages, each with a corresponding conversion rate to measure the proportion of users who continue to the next step in the process. These stages typically include browsing the product page, adding to the cart, proceeding to the checkout page, and completing payment. The conversion rate refers to the proportion of users who successfully complete a stage out of the total number of users who enter that stage, measuring the conversion efficiency of the user at that stage. The number of converted users is the number of users who successfully complete the action at a certain stage. For example, in the "complete payment" stage, the number of converted users is the number of users who ultimately successfully made a payment. The total number of users is the total number of users who entered a certain stage. For example, in the "browse product page" stage, the total number of users is the number of all users who browsed the product page.

[0233] A / B testing was conducted on different product information display methods to evaluate the impact of different information sections on conversion rates.

[0234] ;

[0235] A / B testing is a method that compares the effects of two or more different design schemes to determine which scheme is better. In product information display, A / B testing is used to evaluate the impact of different display methods on user behavior (such as conversion rate). Specifically, it includes: S1. Defining the test objective, i.e., clarifying the goals you hope to achieve through A / B testing, such as increasing purchase conversion rate or increasing user dwell time; S2. Selecting test variables, i.e., determining the product information display method to be tested, for example, display method A places the "user reviews" section at the top of the product details page, while display method B places the "user reviews" section at the bottom of the product details page; S3. Creating a test group and a control group, randomly assigning users to these groups to ensure statistical comparability, for example, the test group uses display method A, and the control group uses display method B; S4. Conducting the test, i.e., showing the test group and control group different product information display methods within the same time period, collecting user behavior data, including browsing time, click count, purchase conversion rate, etc.; S5. Calculating the conversion rate for each group of users, using the formula... S6. Compare the conversion rates of the test group and the control group to evaluate the effectiveness of display method A and display method B; S7. Based on the test results, select the better-performing display method as the final solution.

[0236] For each product information section Calculate its overall score:

[0237] ;

[0238] wherein, , , , , are the weights assigned to each analysis result according to business requirements, and satisfy , The specific formula of is:

[0239] ;

[0240] wherein, and are weight parameters for balancing the influence of keyword matching degree and theme similarity, The specific formula of is:

[0241] ;

[0242] wherein, is an indicator function, which takes the value of 1 if the product information part contains high-frequency keywords, and 0 otherwise;

[0243] The specific formula of is:

[0244] ;

[0245] wherein, is the term frequency vector of the product information part , is the term frequency vector of the theme , is the dot product of vectors and , which is used to measure the similarity of two vectors in direction, is the modulus product of vectors and , which is used for normalization to ensure that the similarity value is between -1 and 1;

[0246] According to the comprehensive score, the key information part is determined as:

[0247] ;

[0248] wherein, is the weight threshold value for determining the key information part. If the comprehensive score of the product information part exceeds the weight threshold value, it is determined as the key information part.

[0249] By employing the methods described above, and through comprehensive analysis of user behavior data and purchase conversion data, we can scientifically determine whether there are any missing product information items. This avoids subjective assumptions, ensures the effectiveness of supplementary measures, and accurately pinpoints the specific missing information, providing a clear direction for subsequent supplementary measures and preventing blind supplementation. By supplementing and improving product information, we can meet users' needs for product information, reduce user confusion and hesitation when browsing product pages, and improve user satisfaction. By optimizing product information display, we can reduce user confusion and hesitation during the purchase process, thereby increasing the purchase conversion rate. Through dynamic adjustments, we can continuously optimize product information display to ensure that the information display effect is always at its best, adapting to market changes and dynamic changes in user needs.

[0250] Example 2 is an improvement upon Example 1. This semantically related automatic product information completion method completes the product information of the target product, including analyzing the user profile of the target product, specifically:

[0251] Get the browsing user set ;

[0252] Extract browsing user set Each user in User profile data includes basic information such as age, gender, region, and consumption level, as well as shopping-related information such as purchase preferences, browsing history, and review history. ;

[0253] in, Indicates user age, Indicates user gender, Indicates user area, Indicates user consumption level Indicates user Purchase preferences Indicates user Browsing history Indicates user Historical evaluation;

[0254] Based on the browsing user set Each user in User profile data is used to segment users, that is, to divide users into different groups: ;

[0255] in, For the number of clusters, Represents an interval, Indicates the first an age interval;

[0256] For each user group, the attention degree of each part of the commodity information is calculated, and the attention points and demands of the user group on the commodity information are analyzed: ; wherein, is the number of users of the user group , represents the i-th part of the commodity information, is the total number of parts of the commodity information, represents the attention degree of the user to the commodity information part , and the specific formula is: ; wherein, represents the dwell time of the user in the commodity information part , represents the number of clicks of the user in the commodity information part , represents the scroll depth of the user in the commodity information part , i.e. the percentage of page scrolling, , , , respectively are the weights of the dwell time, the number of clicks and the scroll depth, and , , , , are used to balance the influence of behavior characteristics such as dwell time, number of clicks and scroll depth on interest degree weight, and by adjusting these weights, the importance of certain behavior characteristics can be emphasized;

[0257] According to the attention degree of each user group, the supplemented commodity information of each part of the commodity information is generated, denoted as :

[0258] ;

[0259] wherein, is an attention degree threshold value, used to determine whether the attention degree of the user to a certain commodity information part is high enough, when the attention degree is greater than the threshold value, it means that the attention degree of the user to a certain commodity information part is high, otherwise it means that the attention degree of the user to a certain commodity information part is low, represents the detailed commodity information supplemented for the user group with high attention degree, represents the default content of the commodity information.

[0260] The embodiment also provides that the commodity information of the target commodity is completed, and further comprises constructing and analyzing a user behavior model, specifically:

[0261] ​Acquiring a set of browsing users ;

[0262] Extracting a set of browsing users Each user in the set of browsing users Behavior data when browsing the target commodity page, including browsing time, click times, scroll depth, browsing path, etc.

[0263] For each user , define its behavior feature vector : ;

[0264] Wherein, represents the th behavior feature of the user , such as browsing time, click times, etc.

[0265] For each user , construct a behavior path matrix : ; wherein, represents the behavior probability of the user from the commodity information part to the next part, is the th part of the commodity information, is the number of times the user jumps from , is the total access times of the user in ;

[0266] For each commodity information part , calculate its access frequency : ; wherein, represents whether the user has accessed the commodity information part , wherein 1 indicates that if the user has accessed the commodity information part , 1 is output, and if the user has not accessed the commodity information part , 0 is output.

[0267] Find out the key path that appears more frequently in the browsing behavior of the user on the target commodity , the key path is a set of frequently appearing behavior paths, which is identified by the frequency of path appearance:

[0268] ; wherein, is the path, For the frequency of path occurrence, A threshold for the frequency of path occurrence is used to filter the set of critical paths. This involves identifying which paths are sufficiently important in user browsing behavior to be considered critical paths.

[0269] For each product information section Calculate its information demand degree Taking into account both node access frequency and behavioral characteristics on the critical path: ;in, For path importance, it represents the importance of the path within the set of critical paths. In the text, the degree of impact of each path on user browsing behavior is as follows: ;in, set of critical paths The path access frequency of a certain path, that is, the number of times a certain path is accessed by users. The total number of paths, i.e., the set of browsing users. The total number of browsing paths for all users in the region. set of critical paths The path length of a certain path, that is, the number of product information segments contained in that path. The average path length, i.e., the number of browsing users. The average length of the browsing path for all users in the system;

[0270] Define a gap detection function to determine the product information section. Is there an information gap?

[0271] ;in, Product information section The level of detail; if Then determine the product information section. There is an information gap;

[0272] like Then determine the product information section. There is no information gap;

[0273] Based on the analysis results of each part of the product information under the gap identification function, supplementary product information is generated for each part of the product information, denoted as... :

[0274] ;

[0275] in, For detailed information to fill the information gaps, This is the default content for product information.

[0276] This embodiment also provides methods for completing product information for the target product, including mining user behavior association rules, specifically:

[0277] Get the browsing user set ;

[0278] Extract browsing user set Each user in Behavioral data while browsing the target product page, including the sequence of product information sections visited;

[0279] For each user Define its sequence of behaviors: ;in, Indicates user The product information section is accessed. The number of product information sections accessed by the user;

[0280] Each user Browsing behavior is treated as a transaction, and a behavior transaction database is built. :

[0281] ; among them, each transaction This represents a set of product information accessed by a user, i.e., each transaction. A sequence of user behaviors is represented as ;

[0282] Use association rule mining algorithms, such as the Apriori algorithm, to mine behavioral transaction databases. Frequent itemsets in: ;in, The minimum support threshold is used to filter frequent itemsets. This threshold sets a standard to distinguish which itemsets appear frequently enough in user behavior data. Only itemsets with a frequency exceeding this threshold are retained as frequent itemsets for further analysis. By setting this threshold, combinations of product information that are rarely accessed by users simultaneously can be filtered out, thereby reducing the interference of noisy data on the analysis. This allows subsequent association rule mining to focus more on meaningful patterns and reduces the interference of unimportant itemsets. It appears more frequently than the minimum support threshold in all transactions in the behavioral transaction database. itemsets, The support function is the function that includes itemsets. The percentage of transactions in the total number of transactions is given by the following formula: ;in, For behavioral transaction database The number of transactions in the middle contains The number of transactions, The total number of transactions in the behavior transaction database ;

[0283] The frequent item sets are generated into association rules in the form of , wherein, and are subsets of the product information part, and , that is, and are disjoint item sets;

[0284] The confidence of the association rules is defined as : ; wherein, is the minimum confidence threshold for evaluating the reliability of the association rules, so the association rules need to meet the minimum confidence threshold ;

[0285] For the association rules , a gap check function is defined to check whether there is missing information related to the frequent subsequent behavior in the product information, for example, if the association rules show that users often view the applicable scenarios after viewing the product parameters, but the product information lacks detailed description of the applicable scenarios, it is considered that there is a gap in the information: ; wherein, represents the detail level of the part of the product information, is the detail level threshold for judging whether the product information part is detailed enough to decide whether more information needs to be supplemented;

[0286] If , it is determined that there is a gap in the product information part;

[0287] If , it is determined that there is no gap in the product information part;

[0288] According to the association rules and the analysis results of each part of the product information under the gap check function, the supplemented product information of each part of the product information is generated, denoted as :

[0289] ;

[0290] Wherein, indicates the existence of certain data or certain conditions, indicates the absence of certain data or certain conditions, is the detailed information supplemented for the information gap, is the default content of the product information.

[0291] The embodiment also provides that the product information of the target product is completed, and further includes cross-platform data integration and analysis, specifically:

[0292] A multi-channel data source set is set, and the multi-channel data source set contains all platform channels related to the target product: ; wherein each data source contains user behavior data and feedback information, including but not limited to browsing records, purchase records, likes, comments, shares, search keywords, etc.

[0293] The multi-channel data source set and all user behavior data and feedback information are integrated to generate an integrated data set: ;

[0294] A browsing user set is obtained;

[0295] The behavior data of each user in the browsing user set in browsing the target product on different platforms is extracted;

[0296] For each user , a behavior vector on different platforms is defined: ; wherein represents the performance of the user on the th behavior;

[0297] Hot topics of users in the browsing user set are identified;

[0298] A user comment set is extracted;

[0299] A hot topic set is extracted;

[0300] For each user comment , the sentiment tendency thereof is analyzed by a sentiment dictionary or a machine learning model: ; wherein represents that the user comment is a positive comment, represents that the user comment is a neutral comment, represents that the user comment is a negative comment;

[0301] A missing analysis function is defined to determine the missing part in the product information of the target product:

[0302] ​ ;

[0303] in, For users to view product information The specific formula for the partial demand is as follows: ;in, and The weights are respectively the access frequency and the relevance to trending topics, and This is used to balance the importance of visit frequency and relevance to trending topics in demand calculation. The frequency of user access is specifically: ;in, Indicates user Did you access the product information section? Output 1 if the site is visited, and output 0 if it is not visited. Represents a collection of browsing users The total number of users in the demand formula; For relevance to trending topics, specifically: ;in, The number of comments related to trending topics, i.e., within the set of trending topics. In the middle, with the product information section The number of related comments The number of all comments, i.e., the set of user comments. The total number of elements; in the demand formula, This is the sentiment weight, used to adjust the demand level, thereby more accurately reflecting users' actual needs for the product information section. Its specific formula is: ;in, It's a comment The emotional tendency is represented by 1 for positive, 0 for neutral, and -1 for negative. It is an indicator function that represents a comment. Does it involve product information? If comments Product information section If the comment is positive, the output will be 1. Section not involving product information If the output is 0, then the output is 0; in the missing data analysis function, The level of detail used to measure the product information section. The completeness and richness of the content are specifically defined by the following formula: ;in, Product information section The length of the text, such as the number of words or characters. Product information section the number of multimedia contents, such as the number of pictures, the number of videos, etc. and is a weight parameter for balancing the influence of the text length and the number of multimedia contents, satisfying ;

[0304] If , it is determined that there is a missing in the part of the product information of the target product;

[0305] If , it is determined that there is no missing in the part of the product information of the target product;

[0306] According to the analysis result of each part of the product information under the missing analysis function, the supplemented product information of each part of the product information is generated, denoted as :

[0307] ;

[0308] wherein, is the detailed information for supplementing the information gap, is the default content of the product information.

[0309] wherein, the hot topics of the users in the set of browsing users are identified, specifically:

[0310] For each data source , all user comments related to the target product in the data source are obtained to form a set of user comments: ; wherein, represents the th user comment, represents the total number of comments, and each comment contains comment text, comment time, user information, score, and other comment-related information;

[0311] For each comment , a set of keywords is extracted: ; wherein, represents the th keyword in the th comment;

[0312] For the keyword , its TF value is calculated: ;

[0313] For the keyword , its IDF value is calculated: ;

[0314] ​​For keywords , calculate their TF-IDF values: ;

[0315] According to the TF-IDF values, extract high-score keywords as a hot topic set :

[0316] ;

[0317] where, is the threshold value of TF-IDF value, used to distinguish which keywords have higher importance in the text set, by setting a threshold , those keywords with TF-IDF values greater than or equal to the threshold can be filtered out, these keywords are usually considered as the words that have significant contribution to the text content.

[0318] In this embodiment, by comprehensively analyzing user behavior data and purchase conversion data, it is scientifically judged whether the commodity information is missing, subjective speculation is avoided, the effectiveness of the completion measures is ensured, the specific part of the information missing is accurately positioned, the subsequent completion measures are provided with clear direction, blind completion is avoided, the commodity information is supplemented and improved, the demand of users for commodity information is met, the confusion and hesitation of users in browsing the commodity page are reduced, the user satisfaction is improved, the commodity information display is optimized, the confusion and hesitation of users in the purchase process are reduced, so as to improve the purchase conversion rate, through dynamic adjustment, the commodity information display is continuously optimized, the information display effect is always in the best state, and the dynamic change of market change and user demand is adapted.

[0319] Embodiment three, this embodiment is an example of all the above embodiments, that is, the overall algorithm of the present application:

[0320] Take a smart phone sold on Taobao as an example:

[0321] Through the Taobao commodity database or commodity information API, the commodity information of the smart phone is obtained according to the commodity ID, including the commodity name, description, parameters (such as processor model, memory size, screen size, etc.), user evaluation, application scenario, after-sales service and other contents;

[0322] Query the Taobao user behavior database or log to obtain the unique identification of all users who have browsed the smart phone, form a browsing user set, and collect a list of users who have browsed the smart phone;

[0323] For each user, obtain their refined behavior data when browsing the smartphone detail page, such as dwell time, number of clicks on the parameter section and user evaluation section, scroll depth, etc., and analyze and identify that users pay more attention to the parameter and user evaluation sections, but pay less attention to the applicable scenarios and after-sales service sections and have confusion (such as frequently clicking the back button or jumping to other pages). Get the browsing behavior analysis results of each user, and find that many users stay in the parameter and user evaluation sections for a long time and click frequently, while they show confusion behavior in the applicable scenarios and after-sales service sections.

[0324] Collect data of the smartphone during the purchase conversion process, including the cart addition rate, checkout page entry rate, payment success rate, and purchase conversion rate, etc. Calculate the cart addition rate to be 50%, the checkout page entry rate to be 80%, the payment success rate to be 90%, and the purchase conversion rate to be 36% (relatively low).

[0325] Calculate the proportion of users who have confusion points to be 35% (more than the threshold of 30%), analyze the purchase conversion whole process data, and find that the purchase conversion rate is lower than the expected threshold of 40%. Combined with the user browsing behavior data, it is found that users have confusion in the applicable scenarios and after-sales service sections. Compared with the product information of the best-selling smartphones of the same type, it is found that the target smartphone has a large difference in the detailed level of the applicable scenarios and after-sales service sections compared with the best-selling products. It is judged that the product information of the target smartphone is missing, mainly in the applicable scenarios and after-sales service sections.

[0326] Among them, compared with the product information of the best-selling smartphones of the same type, it is found that the target smartphone has a large difference in the detailed level of the applicable scenarios and after-sales service sections compared with the best-selling products. Specifically:

[0327] Assuming we have selected three similar best-selling smartphones (S1, S2, S3) for comparison, their information in the applicable scenarios and after-sales service sections are as follows: S1's applicable scenarios content lists in detail the scenarios suitable for gaming, photography, office work, etc., and provides case explanations; S1's after-sales service content provides a 2-year warranty, detailing warranty scope, customer service response time, repair service process, etc.; S2's applicable scenarios content covers various usage scenarios, including gaming, travel, and daily office work, and emphasizes its advantages; S2's after-sales service content includes a 1-year warranty and a worry-free return and exchange policy, detailing the return and exchange process and service network information; S3's applicable scenarios content specifically explains the scenarios suitable for gaming, video editing, and multitasking, accompanied by user scenario stories; S3's after-sales service content provides a 3-year warranty, clearly stating warranty terms, fast response service, and global warranty information; the target smartphone (denoted as T) has simple applicable scenarios content mentioning that it is suitable for various scenarios but does not expand on the details, and its after-sales service content only states that it provides a 1-year warranty without specific details. In the difference calculation, the applicable scenario difference is that, compared to S1, T does not provide case explanations and the content is brief, compared to S2, T does not emphasize the specific advantages of each scenario, and compared to S3, T lacks user scenario stories. The difference summary is that T lacks details and specific cases in the applicable scenario description, and the after-sales service difference is that, compared to S1, T has a shorter warranty period and does not clearly state warranty scope, compared to S2, T does not mention the worry-free return and exchange policy or service network information, and compared to S3, T has a shorter warranty period and does not provide fast response service or global warranty information. The difference summary is that T lacks specific policies and service details in the after-sales service section. The differences between the target product and each best-selling product of the same type will be quantified, with a value of 1 indicating a difference and 0 indicating the same. The difference degree of the target product compared to each best-selling product of the same type is calculated: where m = 2, i.e., including the applicable scenarios and after-sales service two key information parts, for S1, for S2, for S3, then the final difference degree is Assuming the threshold value is set to 0.3, this difference degree is much higher than the threshold value, indicating that the target smartphone has a larger difference in the detailed level of applicable scenarios and after-sales service compared to the best-selling products;

[0328] Combined with user portrait analysis: analyzing the user portrait of those who purchased the smartphone, it is found that the young user group pays more attention to the gaming performance and photography effect of the phone, while business people pay more attention to the adaptability of office applications and data security;

[0329] Building user behavior model: based on user behavior data to build model, find user browsing parameters part after often see user evaluation, but in the applicable scene and after-sales service part of the browsing less and confusion;

[0330] Mining user behavior association rules: through association rule mining, find that users are more interested in the applicable scene after viewing the parameter part, but the current applicable scene part information is insufficient;

[0331] Cross-platform data integration and analysis: integrate user discussions on mobile phones on social media such as microblogging, find that users discuss more about the use experience and after-sales service guarantee of mobile phones in different scenarios;

[0332] Complete the product information of the target product: for the applicable scene part, add detailed introduction and case sharing of mobile phones in game, photography, office and other scenarios, for the after-sales service part, explain the warranty policy, return and exchange process, customer service response time and other information, make the completed smart phone product information more complete, better meet the needs of different user groups, improve user experience;

[0333] According to the feedback of real-time data monitoring system and user behavior analysis results, it is found that users are more interested in the parameter part, so the display position of the parameter part is advanced, and the key parameters are highlighted. For the applicable scene and after-sales service part, after completing the information, through dynamic adjustment of display strategy, increase the corresponding display area and guide mark, so that the adjusted product information display is more in line with user needs, the user stay time is prolonged, and the purchase conversion rate is improved.

Claims

1.A method for automatic completion of commodity information based on semantic association, characterized in that: Comprise: Obtain all commodity information of the target commodity, generate a commodity information part set: ； wherein represents a first part of the product information; Determine whether the commodity information of the target commodity is missing; If the commodity information of the target commodity is not missing, do not perform the information completion operation; If the commodity information of the target commodity is missing, complete the commodity information of the target commodity and perform a dynamic adjustment commodity information display strategy; The execution of the dynamic adjustment commodity information display strategy is specifically: For each part of the product information of the target product , calculate its interest degree weight: ； wherein, represents the dwell time of the user in the product information section , represents the click count of the user in the product information section , represents the scroll depth of the user in the product information section , , , are weight coefficients for the dwell time, click count, and scroll depth, respectively, and , represents the dwell time of the user in the product information section , represents the click count of the user in the product information section , represents the scroll depth of the user in the product information section , , , are weight coefficients for the dwell time, click count, and scroll depth, respectively, and , represents the specific product information section that is currently being analyzed or considered for adjustment of the display strategy, represents all other sections in the product information; For each part of the product information of the target product , calculate its confusion degree weight: ； wherein, represents the specific product information part that is currently being analyzed or considered for adjustment of the display strategy, is the focus part for attention when calculating the confusion degree weight, and is used to determine whether this part needs to be adjusted in terms of display mode according to user behavior data, represents all other parts in the product information, which is used as a comparison or normalization reference when calculating the confusion degree weight, that is, participates in the calculation as a reference system, to ensure that the calculated confusion degree weight has a relative meaning and comparability, represents the number of times the user clicks the back button in the product information part , represents the number of times the user jumps the browsing path in the product information part , , are respectively the weights of the number of times the back button is clicked and the number of times the browsing path is jumped, and , represents the number of times the user clicks the back button in the product information part , represents the number of times the user jumps the browsing path in the product information part , , are respectively the weights of the number of times the back button is clicked and the number of times the browsing path is jumped, and ; For each part of the product information of the target product , determine the display content according to the interest degree weight and the confusion degree weight thereof ； wherein, indicates highlighting of the product information part , indicates simplified display of the product information part , indicates display of the original content of the product information part , is an interest weight threshold value for determining whether the product information part needs to be highlighted, is a confusion weight threshold value for determining whether the product information part needs to be simplified or optimized; Determine whether the commodity information of the target commodity is missing, including sequentially analyzing all users browsing the target commodity, specifically: Acquiring a set of browsing users ; For each user in the set of users browsing extract their refined browsing behavior data while browsing the target product page ; defining an attention function , calculating the attention of the user to each of the product information parts , identifying the points of attention of the user to each of the product information parts during the browsing process: ； wherein, denotes the user the dwell time, in the product information section, denotes the user the number of clicks, in the product information section, denotes the user the scroll depth, i.e. the percentage of page scroll, in the product information section, , , are the weights of the dwell time, the number of clicks, the scroll depth, respectively, and , denotes the th user in the set of users browsing, denotes the th section of product information; A confusion degree function is defined , calculates the confusion degree of the user for each product information part , identifies the possible confusion points of the user for each part of the product information during the browsing process: ； wherein, representing the user in the commodity information section the number of times of clicking the return button, representing the user in the commodity information section the number of times of jumping the browsing path, respectively the weight of the number of times of clicking the return button and the number of times of jumping the browsing path, and ;​ A relevance analysis function is set to determine the user's Relevance points when browsing the target product page: ； wherein, is an attention threshold value for determining whether the user's attention to a certain product information part is high enough; Set a confusion analysis function to determine the user's confusion Confusion points that can exist while browsing the target product page: ; wherein, is a threshold of confusion, for determining whether the product information part needs to be simplified or optimized; According to the result of the attention analysis function of each item information part , the user generates an attention point set, denoted as : ; According to the result of the confusion degree analysis function of each item information part , the user generates a confusion point set, denoted as : ; Integrating a set of concerns and a set of confusions to generate a user's browsing behavior analysis result. 2.The method of claim 1, wherein: Determine whether the commodity information of the target commodity is missing, including sequentially analyzing all users browsing the target commodity, specifically: Acquisition database ; An identifier of a target commodity is acquired, and is designated as a target identifier, denoted as ; Extract all records in the database with the target identification, denoted as ; Define a record analysis function to determine whether the target commodity has browsing records: ； wherein, represents an empty set, i.e. the target identifier has no record in the database, indicating that the target commodity can have no browsing record, or the browsing record of the target commodity is recorded in the user behavior log; If then it is determined that the target product has a browsing record; Then query the corresponding user behavior database of the target identifier in the database to obtain all records of the target commodity in the user behavior database, forming a browsing record set: ； wherein, is a query function for querying the corresponding user behavior database of the target identity in the database, is the user behavior database corresponding to the target identity; Generate a browsing user set: ； in, Indicates user For target products Browsing behavior data, if user Browsed target products If the value is greater than or equal to 1, then the value is 0; otherwise, it is 0. If , it is determined that the target product has no browsing record. Then the query target identification corresponds to the user behavior log in the database , and a query result set is generated ； wherein, is a query function for querying the corresponding user behavior log of the target identifier in the database; Set a query analysis function to determine whether a browsing user set can be generated: ； If then it is determined that the set of browsing users can be generated; If then it is determined that the set of browsing users cannot be generated; Generate a browsing user set: ； in, Indicates user For target products Browsing behavior data, if user Browsed target products If the value is greater than or equal to 1, then it is 0; otherwise, it is 0. This indicates an empty set, meaning that the target identifier has no browsing history and is not recorded in the user behavior log, indicating that the target product has no browsing history. 3.The method of claim 2, wherein: Determine whether the commodity information of the target commodity is missing, including comprehensively analyzing user browsing behavior data and purchase conversion whole process data to determine whether the commodity information of the target commodity is missing, specifically: Acquiring a set of browsing users ; extracting a set of users who browse the target commodity into a shopping cart, generating a set of users who add commodities, denoted as ; calculating the add-to-basket rate of the target commodity, denoted as : ; wherein, is the number of elements of the set of users who added to cart, is the number of elements of the set of users who viewed, is the number of elements of the set of users who viewed. Extracting the users in the add-to-cart user set who enter the settlement page to generate a settlement user set, denoted as ; The settlement rate of the target commodity is calculated, denoted as : ; wherein, is the number of elements of the settlement user set; Extracting the users in the settlement user set who successfully complete payment, generating a payment user set, denoted as ; calculating a payment rate of the target commodity, denoted as : ; wherein, is the number of elements of the payment user set; calculating a purchase conversion rate of the target commodity, denoted as : ; Statistically browse the user set There is a set of confusion points in the middle The number of users, calculate the proportion of confused users : ; wherein, is an indicator function, which takes the value 1 if the user has a set of confusion points which is not empty, i.e. the user has a set of confusion points , and 0 otherwise, is the number of browses; counting the frequency of occurrence of points of confusion of users in the key information section : ； wherein, is the number of key parts of the target commodity commodity information, the number of key parts refers to the number of key information parts determined in the target commodity commodity information, i.e. the number of elements in the set of key information parts , which reflects the number of parts in the commodity information that need to be focused on and optimized, represents is a key information part, is an indicator function, if the user's confusion point set contains the key information part , i.e. the confusion point set and the key information part overlap, then the value is 1, otherwise the value is 0; Obtain the same kind of best-selling goods set, marked as : ; calculating the difference of the target product from the best-selling products of the same kind in each part of the product information : ； wherein, indicates the content of the target product in the product information section , indicates the content of the best-selling product of the same type in the product information section , is an indicator function, which takes the value 1 if the content of the target product in the product information section is different from the content of the best-selling product of the same type , and 0 otherwise, is the number of best-selling products of the same type. Define a comprehensive analysis function to comprehensively determine whether the commodity information of the target commodity is missing: ； wherein, is a confusion user proportion threshold value, used to measure whether the proportion of users who are confused about the product information among users who browse the target product page is too high, is a purchase conversion rate threshold value, used to determine whether the proportion of users who ultimately purchase the target product among users who browse the product page is too low, is a confusion point occurrence frequency threshold value, used to evaluate whether the frequency of occurrence of confusion points when users browse the key information part is too high, is a difference degree threshold value, used to compare the difference degree of product information between the target product and the best-selling product of the same type; If then it is determined that the product information of the target product is missing; If then it is determined that the product information of the target product does not have a missing. 4.The method of claim 1, wherein: Complete the commodity information of the target commodity, including analyzing the user portrait of the target commodity, specifically: Acquiring a set of browsing users ; extracting user profile data for each user in the set of users browsing​ ； in, Indicates user age, Indicates user gender, Indicates user area, Indicates user consumption level Indicates user Purchase preferences Indicates user Browsing history Indicates user Historical evaluation; According to user profile data of each user in the set of users browsing grouping the users:​ ； wherein, is the number of clusters, denotes an interval, denotes the first age interval; For each user group, calculate its attention to each part of the commodity information, analyze its attention points and needs for the commodity information : ； wherein, the number of users in the user group, the number of users, the i-th part of the product information, the total number of parts of the product information, the attention of the user to the part of the product information, is specifically formulated as: wherein, the time of the user staying in the part of the product information, the number of clicks of the user on the part of the product information, the number of clicks of the user on the part of the product information, the scroll depth of the user on the part of the product information, i.e. the percentage of page scrolling, , , , , , the weight of the time of staying, the number of clicks and the scroll depth, respectively, and According to the attention degree of each user group, the commodity information of each part is generated, and the generated commodity information is denoted as : ； wherein, is a threshold of attention, for judging whether the attention of a user to a certain product information part is high enough, represents detailed product information supplemented for a user group with high attention, represents a default content of product information. 5.The method of claim 4, wherein: Complete the commodity information of the target commodity, also including building and analyzing a user behavior model, specifically: Acquiring a set of browsing users ; extracting a set of users who have browsed each user in the set of users who have browsed behavioral data while browsing the target item page; For each user , a behavior feature vector is defined: ； wherein, representing a user 's first behavioral characteristic; For each user , construct a behavior path matrix : ； wherein, representing a user from the product information section the probability of the behavior of jumping to the next section, for the first section of product information, for the user from the number of jumps, for the user the total number of accesses in ; For each item information part , the frequency of its access is calculated : ； wherein, indicates whether the user has accessed the product information section , wherein 1 indicates that the user has accessed the product information section and 0 indicates that the user has not accessed the product information section ; Finding out the key path in which the user appears frequently in the browsing behavior for the target product : ； wherein, is a path, is a path frequency, is a threshold for path frequency, used to filter the set of critical paths determines which paths are important enough in the user's browsing behavior to be considered critical paths; For each commodity information part , calculate its information demand degree , comprehensively consider the node access frequency and the behavior characteristics on the key path: ； wherein, is a path importance, representing the influence degree of each path on the user browsing behavior in the critical path set , specifically: ； wherein, the set of critical paths the path access frequency of a path in the set of critical paths, i.e. the number of times a path is accessed by a user, the total number of paths, i.e. the set of browsing users the total number of paths browsed by all users in the set of browsing users, the set of critical paths the path length of a path in the set of critical paths, i.e. the number of product information parts a path contains, the average path length, i.e. the average length of all paths browsed by the set of browsing users the average length of all paths browsed by all users in the set of browsing users; Define a gap identification function to determine whether the product information section has a gap in information: ； wherein indicates the level of detail of the product information section indicates the level of detail of the product information section If then determine that the product information section has a gap in information; If then determine that the merchandise information portion is free of information gaps; According to the analysis result of each part of the commodity information under the gap identification function, the supplemented commodity information is generated for each part of the commodity information, denoted as : ； wherein, is detailed information for the information gap fill, is the default content of the product information. 6.The method of claim 4, wherein: Complete the commodity information of the target commodity, also including mining user behavior association rules, specifically: Acquiring a set of browsing users ; extracting a set of users who have browsed each user in the set of users who have browsed behavioral data while browsing the target item page; For each user , define its sequence of actions: ； wherein, represents the user accessed the product information section, is the number of product information sections accessed by the user; Each user Browsing behavior is treated as a transaction, and a behavior transaction database is built. : ； wherein each transaction represents a set of product information sections accessed by a user, denoted as ; Mining frequent itemsets in behavior transaction databases : ； wherein, is a minimum support threshold for filtering frequent itemsets, is a support function, i.e. the proportion of the number of transactions containing the itemset The frequent item set generation form is an association rule; Defining the confidence of an association rule : ； wherein, is a minimum confidence threshold for assessing the reliability of the association rules, so the association rules need to satisfy the minimum confidence threshold ; For association rules , a gap check function is defined to check whether there is missing information related to the frequent subsequent behavior in the commodity information: ； wherein, indicates the level of detail of the product information part indicates the level of detail of the product information part is a threshold value for the level of detail, for determining whether the product information part is detailed enough, so as to decide whether more information is needed or not. If then it is determined that there is a gap in the information of the product information part; If then it is determined that the product information has no information gap in this part; According to the association rule and the analysis result of each part of the product information under the gap check function, the product information after the supplement is generated for each part of the product information, denoted as : ； wherein, represents that there is certain data or certain condition, represents that there is no certain data or certain condition, is detailed information for supplement of information gap, is a default content of product information. 7.The method of claim 4, wherein: Complete the commodity information of the target commodity, also including cross-platform data integration and analysis, specifically: Set a multi-channel data source set: ； wherein each data source contains user behavior data and feedback information; Integrate the multi-channel data source set and all user behavior data and feedback information to generate an integrated data set: ； Acquiring a set of browsing users ; extracting browsing user set each user in the set behavior data of browsing target goods on different platforms; For each user , define its behavior vector on different platforms: ； wherein representing a user In a first behavioral performance; Identifying a set of browsing users Hot topics among users; Extracting a set of user reviews ; Extracting a set of hot topics ; For each user review , analyze its sentiment orientation : ； wherein, represents a positive review belongs to a positive review, represents a user review belongs to a neutral review, represents a user review belongs to a negative review; Define a missing analysis function to determine the missing part of the commodity information of the target commodity: ； wherein, is a demand degree of the user for the product information part, is a demand degree of the user for the product information part, is a detail degree, used to measure the content integrity and richness of the product information part, is a detail degree, used to measure the content integrity and richness of the product information part, If , it is determined that there is a missing part in the product information of the target product . If , it is determined that there is no absence in the product information of the target product in the part. According to the analysis result of each part of the commodity information under the missing analysis function, the commodity information of each part of the commodity information is generated after being supplemented, denoted as : ； wherein, is detailed information for the information gap, is the default content of the product information. 8.The method of claim 7, wherein: Identifying a set of browsing users Hot topics among users, in particular: For each data source , obtain all user reviews in the data source related to the target product, forming a user review set: ； wherein, represents the user comment, represents the total number of comments; For each review , extract a set of keywords : ； wherein, represents the th keyword in the th comment; For the keyword , its TF value is calculated: ； For the keyword , the IDF value is calculated: ； For the keyword , its TF-IDF value is calculated: ； According to the TF-IDF value, extract high-scored keywords as a hot topic set : 。

Images