A Smart Product Retrieval and Recommendation Method for Industry Databases

Abstract

This invention discloses a method for intelligent product retrieval and recommendation based on an industry database. The invention relates to the field of data processing technology. It obtains user-inputted search content through an interactive interface, extracts product keywords, and performs a fuzzy search in an industry database to generate a candidate product set. Further, it extracts the click-through rate, viewing time, and conversion rate of candidate products from search logs, constructs a search matching model to calculate a search matching index, and generates a search product sequence to be displayed to the user. Simultaneously, it extracts priority products from the search product sequence and, based on the priority product's sequence number, extracts recommended products from a product association table to generate a recommended product sequence, which is displayed to the user synchronously with the search product sequence to achieve accurate product recommendations.

Description

Technical Field

[0001] This invention belongs to the field of data processing technology, specifically, it relates to a method for intelligent product retrieval and recommendation for industry databases. Background Technology

[0002] With rapid economic development and an increasingly diverse range of goods, the product information in industry databases is also growing explosively. Consumers are finding it increasingly difficult to find the products they need from this massive amount of information, making the demand for product retrieval and recommendation systems ever more urgent.

[0003] Product retrieval and recommendation methods targeting industry databases typically rely on simple keyword matching and basic statistical indicators, lacking in-depth analysis of user behavior and multi-dimensional data fusion. This results in significant drawbacks in the accuracy and user experience of search results and recommendations. Specifically, existing technologies often directly match user-input keywords with product names in the database, ignoring the comprehensive influence of users' historical behavior data. This makes the search result ranking unintelligent, failing to prioritize highly relevant or high-conversion-potential products. Users need to spend more time filtering irrelevant or low-quality options, reducing search efficiency. Furthermore, the recommendation modules of existing technologies are usually independent of the search process, using general rules to generate recommendation lists without considering the priority product sequence in the current search context. This leads to low matching between recommended products and users' actual needs, easily resulting in irrelevant recommendations that affect users' purchasing decisions. In terms of data processing, existing technologies lack dynamic optimization of recommended products and cannot respond to changes in user preferences in real time.

[0004] To address the aforementioned issues, this invention proposes a method for intelligent product retrieval and recommendation based on industry databases. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides an intelligent product retrieval and recommendation method for industry databases, solving the problems of inaccurate retrieval and irrelevant recommendations in existing technologies.

[0006] The objective of this invention can be achieved through the following technical solutions:

[0007] A method for intelligent product retrieval and recommendation based on industry databases, the method comprising:

[0008] Step 1: Obtain the search content input by the user based on the interactive interface, extract the product keywords from the search content, perform the search operation based on the product keywords and a pre-built industry database, determine all candidate products that match the product keywords, and generate a set of candidate products;

[0009] Step 2: Obtain the search logs associated with all candidate products in the candidate product set that match the product keywords from the industry database. Extract the click-through rate, viewing time, and conversion rate associated with each candidate product under the search logs of the corresponding product keywords. Build a search matching model to calculate the search matching index between each candidate product and the product keywords. Use the determined search matching index combined with the candidate product set to generate a search product sequence and display it to the user.

[0010] Step 3: Extract priority products from the searched product sequence. Based on the serial number of the priority products in the searched product sequence, extract the number of recommended products associated with the serial number of the corresponding priority products from the industry database. Generate a recommended product sequence based on the serial number of each priority product and display it to the user synchronously with the searched product sequence.

[0011] As a further aspect of the present invention, in step one, the specific method for obtaining the user-input search content based on the interactive interface and extracting product keywords from the search content is as follows:

[0012] Obtain the interaction interface for user interaction, and obtain the search content input by the user based on the interaction interface;

[0013] Based on the content to be retrieved, a set of vocabulary units is determined, which includes single-character units, two-character combinations, and complete noun phrases;

[0014] Obtain a pre-built product core vocabulary table and modifier filter table from the industry database. The product core vocabulary table contains industry standard product names, and the modifier filter table contains descriptive terms.

[0015] A modifier filter table is used to filter and remove descriptive words from the vocabulary unit set;

[0016] All vocabulary units in the filtered vocabulary unit set are matched with the product core vocabulary list to determine all product vocabulary. Then, according to the order of the product vocabulary in the search content, they are combined into a product vocabulary set {A1,A2,...,Aj}, where j is the total number of product vocabulary.

[0017] All product terms in the product terminology set are considered as product keywords in the content to be searched.

[0018] As a further aspect of the present invention, the specific method for determining all candidate products that match the product keywords and generating a candidate product set in step one is as follows:

[0019] Retrieve product inventory tables from industry databases;

[0020] Extract any one product keyword Ai from the product vocabulary set {A1,A2,...,Aj}, where i is the counting index, and its value ranges from 1 to j;

[0021] Perform a fuzzy search query on the product inventory table for the product keyword Ai, extract all products that match the product keyword Ai and have an inventory of not less than 1, as candidate products, and randomly combine the product IDs of all candidate products to generate a set of candidate products associated with the product keyword Ai, where the product ID is unique;

[0022] Similarly, determine the candidate product set associated with all product keywords in the product vocabulary set {A1,A2,...,Aj}.

[0023] As a further aspect of the present invention, in step two, the specific method for extracting the click-through rate, browsing time, and conversion rate associated with each candidate product under the search log of the corresponding product keywords is as follows:

[0024] Extract the set of candidate products associated with any given product keyword Ai;

[0025] Retrieve the retrieval log table from the industry database. The retrieval log table contains product ID, session ID, behavior type, behavior duration, and transaction flag.

[0026] Behavior types include click behavior;

[0027] Behavior duration refers to the time a user spends browsing a product after clicking on it;

[0028] The transaction indicator shows whether the user has completed the purchase after clicking on the product. Yes means the purchase is complete, no means the purchase is not complete.

[0029] The set of candidate products associated with the product keyword Ai is represented as {a1,a2,...,am}, where m represents the total number of candidate products;

[0030] Extract the product ID of any candidate product from the candidate product set {a1,a2,...,am}: an, where n is the counting index, with a value ranging from 1 to m;

[0031] For an, perform the following steps:

[0032] Step 1: Count the number of unique session IDs with the keyword "Ai" in the search log table, and use this as the total number of impressions S;

[0033] Step 2: Count the number of records in the search log table where the keyword for the queried product is Ai, the product ID is an, and the behavior type is click. Use these records as the click count C_click for the candidate product with product ID an.

[0034] Step 3: Calculate the click-through rate (CTR). The formula is CTR = C_click / S. If S = 0, set CTR = 0.

[0035] Step 4: Extract all records from the retrieval log table where the query product keyword is Ai, the product ID is an, and the behavior type is click. Obtain the behavior duration of all records and take the arithmetic mean as the viewing time T_view. If there are no click records, set T_view=0.

[0036] Step 5: Count the number of records in the retrieval log table where the keyword for the queried product is Ai, the product ID is an, the behavior type is click, and the transaction flag is yes. This number is recorded as the number of transactions, C_purchase.

[0037] Step 6: Calculate the conversion rate CR. The formula is CR = C_purchase / C_click. If C_click = 0, set CR = 0.

[0038] Repeat the above steps to determine the click-through rate, viewing time, and conversion rate of all candidate products in the candidate product set under the search logs of the corresponding product keywords.

[0039] As a further aspect of the present invention, the specific method for constructing the retrieval matching model to calculate the retrieval matching index between each candidate product and product keywords in step two is as follows:

[0040] The click-through rate (CTR), view time (T_view), and conversion rate (CR) of any candidate product with product ID an in the candidate product set {a1,a2,...,am} associated with product keyword Ai;

[0041] Calculate the maximum and minimum lookup times T_max and T_min for all candidate items in the candidate item set {a1,a2,...,am}. Then, use the maximum and minimum values ​​T_max and T_min to calculate the normalized lookup time T_norm associated with the lookup time T_view. The calculation steps are as follows:

[0042] If T_max≠T_min, T_norm=(T_view-T_min) / (T_max-T_min);

[0043] If T_max = T_min, then T_norm = 0.5;

[0044] The retrieval matching index ML associated with the candidate product with product ID an is calculated using ML=ω1×CTR+ω2×T_norm+ω3×CR, where ω1, ω2, and ω3 are all preset weights and satisfy ω1+ω2 +ω3=1, and ω1, ω2, and ω3 all belong to [0,1].

[0045] Repeat the above steps to determine the retrieval matching index of all candidate products in the candidate product set {a1,a2,...,am}.

[0046] As a further aspect of the present invention, the specific method for generating the search product sequence by combining the determined search matching index with the candidate product set in step two is as follows:

[0047] Extract the candidate product set {a1,a2,...,am} associated with the first product keyword A1 in the product vocabulary set {A1,A2,...,Aj};

[0048] Then, all candidate products in the candidate product set {a1,a2,...,am} are arranged in descending order according to their respective search matching indices, and the arranged result is recorded as the search product subsequence associated with product keyword A1;

[0049] Similarly, determine the subsequences of search products associated with all product keywords in the product terminology set {A1,A2,...,Aj}, totaling j subsequences. Use each subsequence as a parameter in the search product sequence, combine the j subsequences, and generate the search product sequence RPS associated with the product terminology set {A1,A2,...,Aj}.

[0050] As a further aspect of the present invention, in step two, after determining the search product sequence RPS, the search product subsequences in the search product sequence RPS are sequentially displayed to the user in a paginated manner in the display section of the interactive interface.

[0051] As a further aspect of the present invention, the specific method for extracting priority products from the retrieved product sequence in step three is as follows:

[0052] Extract the first item from all subsequences of retrieved items from the retrieved item sequence RPS, and arrange them into a priority item sequence a1', a2', ..., aj' according to the order of the retrieved item subsequences.

[0053] As a further aspect of the present invention, the specific method for generating the recommended product sequence in step three is as follows:

[0054] Using the product IDs of each priority product in the priority product sequence a1', a2', ..., aj' as input, the recommended products associated with the sequence numbers of each priority product are extracted from the product association table in the industry database. The product association table records the association relationship and association weight between two products. The association weight is determined by the number of times the two products are purchased by the same user within a preset time period. The more times they are purchased at the same time, the greater the association weight, and vice versa.

[0055] Arrange all recommended products in the order of priority product sequence a1', a2', ..., aj', that is, first arrange all recommended products associated with a1', then arrange all recommended products associated with a2', and so on, to generate a recommended product sequence;

[0056] Among them, the recommended products associated with the priority product with product ID a1' are arranged in a random order, and the other recommended products are arranged in the same way;

[0057] The generated recommended product sequence will be displayed to the user in a pop-up window.

[0058] As a further aspect of the present invention, the specific method for extracting the recommended products associated with the serial numbers of each priority product from the product association table in the industry database in step three is as follows:

[0059] Get the priority item sequence a1',a2',...,aj';

[0060] Determine the weight coefficient of the recommended products associated with each of the j priority products;

[0061] Among them, the weight coefficient of the recommended product for any priority product with product ID ai' is Wi=1 / i;

[0062] The sum of the weight coefficients of the j priority items is W_total=Σ(Wi), i∈[1,j];

[0063] Calculate the percentage of recommended items for any given priority item: Pi = Wi / W_total;

[0064] Get the preset total number of recommended products N_total, and calculate the number of recommended products for any priority product N_total_i=floor(Pi*N_total), where floor is the floor function;

[0065] For any preferred product with product ID ai', extract the top N_total_i products with the highest association weight from the product association table as its associated recommended products;

[0066] Similarly, identify the recommended products associated with all preferred products.

[0067] The beneficial effects of this invention are:

[0068] (1) This invention improves user experience and business efficiency by constructing a complete closed loop from accurate retrieval to intelligent recommendation. First, candidate products are locked in the industry database through keywords, ensuring the professionalism and breadth of the retrieval. Then, an innovative retrieval matching model is introduced to calculate the matching index by combining click-through rate, viewing time and conversion rate, so that the retrieval product sequence presented to users is not only relevant, but also a high-quality choice that has been verified by the market, thus improving the accuracy of the retrieval and conversion potential. Second, the method of this invention does not stop at the retrieval results, but cleverly uses the sequence number association to mine potential recommended products, realizing the synchronization and linkage between retrieval results and recommended content. This design can proactively mine potential needs for users, extending the single retrieval behavior into an efficient personalized shopping guide, which not only saves users time, but also brings additional traffic and transaction opportunities to products through accurate exposure. In the end, while improving user satisfaction, it also enhances the overall service efficiency and business value of the platform.

[0069] (2) This invention obtains user input through an interactive interface and effectively filters irrelevant descriptive words based on a multi-type vocabulary unit combined with a pre-built core vocabulary list and modifier filter list for products, thereby accurately extracting product keywords, which significantly improves the accuracy and relevance of the search and reduces interference. Subsequently, the product inventory table is queried through fuzzy search to ensure that each keyword can match the available candidate products in the inventory and generate a set of candidate products. This not only enhances the robustness of the system and can handle incomplete or biased user input, but also improves the search efficiency and practicality and avoids invalid results. The whole process, from keyword extraction to candidate product generation, is closely linked and optimizes resource utilization.

[0070] (3) This invention systematically extracts key behavioral data such as click-through rate, browsing time and transaction rate from user search logs, constructs a search matching model to calculate the matching index between each candidate product and keyword, and generates a sorted search product sequence based on this and displays it in a paginated form, thereby significantly improving the accuracy and relevance of search results; its advantage is that it makes full use of actual user behavior data to drive the sorting, effectively captures user intent and product quality, and makes recommended products more in line with user needs; at the same time, through normalization processing and weight integration, it optimizes the objectivity and consistency of sorting, and ultimately improves the overall user experience and platform service efficiency;

[0071] (4) This invention achieves accurate and efficient personalized recommendations by extracting priority products from the search product sequence and generating a recommendation sequence based on the product association table. Its core advantage lies in making full use of the user's original search intent and the strong association between products, and dynamically allocating the recommendation quantity of each priority product through weight coefficients to ensure that the core preference products ranked first receive more recommendation weight, thereby improving the relevance and hit rate of the recommendations. At the same time, it combines association weights to select high-frequency co-occurring products and organizes the recommendation sequence according to the search order, which not only strengthens the coherence of the recommendation logic, but also takes into account the diversity of products and the coverage of user interests. Finally, it is presented intuitively in the form of a pop-up window, which effectively improves the user experience and conversion efficiency without hindering the viewing of candidate products. Attached Figure Description

[0072] The invention will now be further described with reference to the accompanying drawings.

[0073] Figure 1 This is a flowchart illustrating the method described in this invention;

[0074] Figure 2 This is a flowchart illustrating the method described in Embodiment 2 of the present invention;

[0075] Figure 3 This is a flowchart illustrating the method described in Embodiment 3 of the present invention. Detailed Implementation

[0076] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0077] Example 1

[0078] A method for intelligent product retrieval and recommendation based on industry databases, such as Figure 1 As shown, this method includes the following:

[0079] This method, an intelligent product retrieval and recommendation method for industry databases, mainly includes the following steps:

[0080] First, the system obtains the user's input search query content through an interactive interface, extracts product keywords from the search query content, performs a search operation based on the product keywords and a pre-built industry database, identifies all candidate products that match the product keywords, and generates a candidate product set. Specifically:

[0081] The interactive interface is an APP or website that interacts with users, through which users can perform operations such as searching, browsing, and purchasing.

[0082] Based on the search content input by the user, keyword extraction is first performed. Natural language processing technology is used to identify core product words from the search content input by the user through word segmentation and entity recognition.

[0083] Then, these core product terms are used to retrieve all relevant products from a pre-built industry database to form a candidate product set. This step reduces query ambiguity and improves the recall rate of the retrieval through semantic analysis, ensuring that all potentially relevant products are taken into consideration.

[0084] Next, the system retrieves search logs associated with all candidate products in the candidate product set matching the product keywords from the industry database. It then extracts the click-through rate, viewing time, and conversion rate associated with each candidate product under the corresponding product keyword search logs. A search matching model is constructed to calculate the search matching index between each candidate product and the product keywords. Using the determined search matching index in conjunction with the candidate product set, a search product sequence is generated and displayed to the user. Specifically:

[0085] Retrieve search logs for each product in the candidate product set from the industry database under the corresponding product keywords. The search logs record users' historical behavior data, including click-through rate, browsing time, and conversion rate.

[0086] Among them, click-through rate reflects users' initial interest, browsing time measures user engagement, and conversion rate measures conversion effectiveness;

[0087] Based on this data, a retrieval matching model is constructed to calculate the retrieval matching index of all candidate products and product keywords. Then, the candidate products are sorted according to the retrieval matching index to generate a retrieval product sequence and display it to the user. The principle of this step is to use user behavior data to optimize the sorting, so that more relevant and popular products are displayed first, thereby improving user experience and retrieval efficiency.

[0088] Finally, priority products are extracted from the searched product sequence. Based on the sequence number of the priority products in the searched product sequence, a number of recommended products associated with the corresponding priority product sequence number are extracted from the industry database. All recommended products are then used to generate a recommended product sequence based on the sequence number of each priority product, and this sequence is displayed to the user synchronously with the searched product sequence. Specifically:

[0089] Extract priority products from the search product sequence. These priority products represent items that users are most likely to be interested in. Based on the sequence number of the priority products in the sequence, extract a number of related recommended products from the industry database.

[0090] For example: For each priority product, the number of recommended products is dynamically determined based on its serial number; the earlier the serial number, the more products are recommended.

[0091] All recommended products are assembled into a recommended product sequence and displayed to users synchronously with the searched product sequence. By using similar product recommendations and popularity bias strategies, users are guided to discover more related products through priority products, increasing browsing depth and sales opportunities, thereby achieving personalized recommendations and sales optimization.

[0092] Example 2

[0093] This embodiment further discloses a method for generating a candidate product set based on embodiment 1, such as... Figure 2 As shown, it specifically includes the following:

[0094] Based on the content described in Example 1, the content to be searched input by the user can be obtained through the interactive interface that interacts with the user;

[0095] Next, the retrieved content is segmented into words. Based on semantic analysis, a set of lexical units is finally obtained. It should be explained that the set of lexical units includes single-character units, two-character combinations, and complete noun phrases. This can be fully achieved based on existing semantic databases, so this solution has not been elaborated on in detail.

[0096] This approach ensures that all possible word units are captured to the greatest extent possible, regardless of whether the user input is long or short, formal or colloquial, thus avoiding the omission of potential matches.

[0097] Then, the core product vocabulary list and the modifier filter list are obtained from the industry database pre-built by the operators. The core product vocabulary list contains industry standard product names, and the modifier filter list contains descriptive words.

[0098] For example: the core vocabulary list for products includes: "T-shirt", "dress", etc.;

[0099] The modifier filter table includes words such as "red", "large", "cheap", and "best".

[0100] The obtained modifier filter table is used to filter and remove descriptive words in the vocabulary unit set. The purpose is to focus on identifying core product entities and avoid these words interfering with the retrieval of core products, so that subsequent retrieval and recommendation operations can be performed only on core product words.

[0101] Then, all the word units in the filtered word unit set are matched with the product core vocabulary list to determine all product words. The matching method is also semantic matching, which can be achieved based on existing technology.

[0102] Then, all the determined product terms are combined into a product term set {A1,A2,...,Aj} according to the order of the product terms in the search content, where j is the total number of product terms. This operation of combining them according to the order of the search content preserves the contextual information of the user's query.

[0103] It should be explained that all product terms in the product terminology set are considered as product keywords in the content to be searched. In the following text, all product terms in the product terminology set will be expressed as product keywords.

[0104] Next, retrieve the product inventory table from the industry database;

[0105] Extract the product vocabulary set {A1,A2,...,Aj} determined above, and then extract any product keyword from the product vocabulary set {A1,A2,...,Aj}, denoted as Ai, where i is the counting index, with a value range from 1 to j;

[0106] For the extracted product keyword Ai, a fuzzy search query is performed in the product inventory table to extract all products that match the product keyword Ai and have an inventory of not less than 1. Here, exact matching is not used, but the LIKE operator in SQL is used. The purpose of the fuzzy search in the full-text search engine is to handle input problems such as spelling errors, synonyms, and abbreviations, or search content that has not been processed by semantic matching. This is because semantic matching may not be able to handle all word units in the search content. Therefore, a fallback process is performed based on this step. Word units that cannot be processed by semantic matching are directly added to the product vocabulary set in their original form.

[0107] Select products with an inventory of at least 1 unit as candidate products, excluding out-of-stock items from the outset to avoid showing users products that cannot be purchased, thereby improving user experience and conversion rates.

[0108] Then, the product IDs of all candidate products are randomly combined to generate a set of candidate products associated with the product keyword Ai, where the product IDs are unique;

[0109] By repeating the above steps, we can determine the set of candidate products associated with all product keywords in the product vocabulary set {A1,A2,...,Aj}.

[0110] Example 3

[0111] This embodiment further discloses a method for generating a search product sequence based on user search content, building upon embodiment 2. Figure 3 As shown, it specifically includes the following:

[0112] Based on the content described in Example 2, obtain a set of candidate products associated with any product keyword Ai;

[0113] Next, the system retrieves a pre-built search log table from the industry database. This search log table includes product ID, session ID, behavior type, behavior duration, and transaction flag. Specifically:

[0114] The behavior types include click behavior, the behavior duration represents the time the user spends browsing the product after clicking on it, and the transaction indicator represents whether the user has completed the purchase after clicking on the product; yes means the purchase is completed, no means the purchase is not completed.

[0115] Next, the set of candidate products associated with the product keyword Ai is represented as {a1,a2,...,am}, where m represents the total number of candidate products in the set of candidate products associated with the product keyword Ai;

[0116] Next, extract the product ID of any candidate product from the candidate product set {a1,a2,...,am} and mark it as an, where n is the counting index, with a value range from 1 to m. The following steps are to perform example processing on the candidate product with product ID an, and the remaining candidate products are processed in the same way.

[0117] First, count the number of unique session IDs for product queries with the keyword "Ai" from the search log table, and mark the count as the total number of impressions S;

[0118] Next, the number of records in the retrieval log table where the keyword for the queried product is Ai, the product ID is an, and the behavior type is click is counted, and the number of records is used as the click count C_click for the candidate product with product ID an.

[0119] Based on the determined number of clicks C_click and the total number of impressions S, calculate the click-through rate (CTR) of the candidate product with product ID an. The calculation formula is CTR=C_click / S, where if S=0, set CTR=0.

[0120] Click-through rate (CTR) reflects the probability that a user will click on a product after seeing it. A high CTR indicates that the product title, image, or description is highly matched with the keywords and can attract the user's attention.

[0121] Next, extract all records from the retrieval log table where the query product keyword is Ai, the product ID is an, and the behavior type is click behavior. Obtain the behavior duration of all records and take the arithmetic mean of the behavior duration of all extracted records as the viewing time T_view of the candidate product with product ID an. If there are no click records, set T_view=0.

[0122] The viewing time T_view measures the length of time a user spends browsing product details after clicking. A long viewing time usually indicates that the user is highly interested and may be carefully comparing or considering a purchase.

[0123] Then, the number of records in the retrieval log table where the query keyword is Ai, the product ID is an, the behavior type is click, and the transaction flag is yes is counted. This number is taken as the number of transactions C_purchase. The conversion rate CR is calculated by using: CR=C_purchase / C_click. If C_click=0, CR=0.

[0124] CR reflects the percentage of actual purchases after a click. A high CR indicates that the product meets user needs and has strong conversion capabilities.

[0125] Finally, by repeating the above steps and performing the same processing on all candidate products in the candidate product set, it is possible to determine the click-through rate, viewing time, and conversion rate associated with all candidate products in the candidate product set under the search logs of the corresponding product keywords.

[0126] By using multi-dimensional behavioral indicators, we can comprehensively capture the performance of a product from exposure to transaction, avoid the bias of a single indicator, and provide a data foundation for subsequent ranking.

[0127] Next, a retrieval matching model is constructed to calculate the retrieval matching index between each candidate product and the product keywords, as follows:

[0128] First, based on the above difficulties, obtain the click-through rate (CTR), viewing time (T_view), and conversion rate (CR) of any candidate product with ID an from the candidate product set {a1,a2,...,am} associated with any product keyword Ai.

[0129] Based on the method described above for determining the viewing time T_view, calculate the maximum value T_max and minimum value T_min of the viewing time for all candidate products in the candidate product set {a1,a2,...,am};

[0130] Next, the maximum and minimum values ​​of the lookup time T_max and T_min are used to normalize the lookup time T_view, and the normalized lookup time T_norm associated with the lookup time T_view is determined. The calculation formula is divided into two directions.

[0131] First, if T_max ≠ T_min, then the following method is used:

[0132] T_norm=(T_view-T_min) / (T_max-T_min);

[0133] Secondly, if T_max = T_min, then T_norm = 0.5.

[0134] Next, we will use:

[0135] ML=ω1×CTR+ω2×T_norm+ω3×CR

[0136] Calculate the search matching index ML associated with the candidate product with product ID an, where ω1, ω2, and ω3 are all preset weights and satisfy ω1+ω2+ω3=1, and ω1, ω2, and ω3 all belong to [0,1].

[0137] Since the time to view (T_norm) has dimensions, the influence of dimensions is eliminated through normalization. However, the click-through rate (CTR) and conversion rate (CR) are specific percentages, which are numerical values, so they can be calculated directly without considering the dimension issue.

[0138] Finally, by repeating the above steps, the retrieval matching index of all candidate products in the candidate product set {a1,a2,...,am} can be determined;

[0139] Next, extract the candidate product set {a1,a2,...,am} associated with the first product keyword A1 in the product vocabulary set {A1,A2,...,Aj}. The candidate product set {a1,a2,...,am} can represent the candidate product set associated with any product keyword. Here, it is used to represent the candidate product set associated with product keyword A1.

[0140] Then, all candidate products in the candidate product set {a1,a2,...,am} are sorted from largest to smallest according to the value of the retrieval matching index associated with each candidate product, to obtain the retrieval product subsequence associated with product keyword A1;

[0141] Repeat the above steps to determine the search product subsequences associated with each product keyword in the product vocabulary set {A1,A2,...,Aj}, and finally obtain a total of j subsequences;

[0142] Then, using j subsequences as a parameter in the product sequence, the j subsequences are combined to generate the product sequence RPS associated with the product vocabulary set {A1,A2,...,Aj}, where the order of the product sequence RPS is the same as the order of the product vocabulary set {A1,A2,...,Aj}.

[0143] Finally, the search results are displayed to the user in a paginated format in the interactive interface, according to the order of the search product subsequences in the search product sequence RPS.

[0144] Example 4

[0145] This embodiment further discloses a method for generating a recommended product sequence based on embodiment 3, specifically including the following:

[0146] Based on the content described in Example 3, the retrieval product sequence RPS is extracted. The first product in all the retrieval product subsequences in the retrieval product sequence RPS is obtained in sequence, and finally the number of products corresponding to the number of retrieval product subsequences is obtained. The extracted products are arranged according to the order of the retrieval product subsequences and denoted as the priority product sequence a1', a2', ..., aj'. This step is to select the best representative products from the search results of each keyword, thereby forming a concise and high-quality seed product list.

[0147] Then, obtain the total number of recommended products N_total preset by the operator for backup;

[0148] Next, using the product ID of each priority product in the priority product sequence a1', a2', ..., aj' as input, the recommended products associated with the sequence number of each priority product are extracted from the product association table in the industry database. The total number of recommended products associated with all priority products shall not exceed the operator's preset limit of the total number of recommended products N_total.

[0149] It should be noted that the product association table is a pre-built table by the operator to record the association relationship and association weight between any two products. The association weight is determined by the number of times the same user purchases the two products at the same time within a preset time period. The more times they are purchased at the same time, the greater the association weight, and vice versa.

[0150] For example, given product X and product Y, if 500 out of 1000 users purchase product X or product Y and then purchase another product within a specified preset time period (3 days), meaning they bought both products, then the association weight between product X and product Y is equivalent to 0.5. If 800 users purchase both products simultaneously, then the association weight is 0.8. The larger the association weight value, the stronger the association relationship between the two products. This is reflected based on the magnitude of the association weight value.

[0151] Next, all recommended products are arranged in the order of priority product sequence a1', a2', ..., aj'. That is, first arrange all recommended products associated with a1', then arrange all recommended products associated with a2', and so on, to finally generate a recommended product sequence.

[0152] It should be noted that the order of all recommended products associated with the priority product with product ID a1' is randomized, and the same applies to all recommended products associated with the other priority products;

[0153] Finally, the generated recommended product sequence is displayed to the user in a pop-up window.

[0154] It should be added that, in the step of determining the recommended products associated with the serial numbers of each priority product, the parameter of "associated quantity" needs to be determined through the following steps;

[0155] First, we need to obtain the priority product sequence a1',a2',...,aj';

[0156] Then, the weight coefficients of the recommended products associated with each of the j priority products are determined, as follows:

[0157] First, calculate the weight coefficient of the recommended products for any priority product with product ID ai' as Wi=1 / i. Wi=1 / i is equivalent to a decreasing weight model. The principle is that the priority products (such as a1') in the priority product sequence a1', a2', ..., aj' correspond to the more core and earlier keywords entered by the user in the search intent, and are therefore considered more important and should receive more recommendation resources.

[0158] Then, the total weight coefficients of the j priority items are calculated as W_total=Σ(Wi), i∈[1,j], that is, W1 to Wj are summarized;

[0159] The percentage of recommended items associated with each priority item is calculated using Pi = Wi / W_total;

[0160] Finally, the recommended quantity N_total_i of any priority product is calculated by using N_total_i=floor(Pi*N_total), where floor is the floor function. For example, if N_total_i=10.9, then N_total_i=10.

[0161] Next, for any preferred product with product ID ai', the top N_total_i products with the highest association weight can be extracted from the product association table as recommended products associated with the preferred product with product ID ai'.

[0162] All data in the formulas described above are numerical calculations performed after removing their dimensions. Furthermore, any content not described in detail in this specification is existing technology known to those skilled in the art.

[0163] The above description is merely an example and illustration of the present invention. Those skilled in the art can make various modifications or additions to the specific embodiments described, or use similar methods to replace them, as long as they do not deviate from the invention or exceed the scope defined in the claims, all of which should fall within the protection scope of the present invention.

[0164] It should be stated that all user data collected in this application was collected with the user's consent and authorization. Furthermore, the uses of user data are legal and compliant, and the use and processing of user data comply with the relevant laws, regulations, and standards of the relevant regions.

Claims

1. An industry-oriented database commodity intelligent retrieval and recommendation method, characterized in that, The method comprises: Step 1: based on the interactive interface, obtaining the user inputted content to be searched, and extracting the product keywords in the content to be searched: A1, A2,..., Aj, wherein j represents the total number of product keywords; Based on the product keywords, the search operation is performed on the pre-constructed industry database to determine all candidate products that meet the product keywords and generate a candidate product set; Step 2: obtaining all candidate products associated with the search log from the industry database in the candidate product set that meets the product keywords, and extracting the click rate, viewing time and transaction rate associated with each candidate product under the search log of the corresponding product keyword in the following specific manner: Extract the candidate product set associated with any one product keyword Ai; Obtain the search log table from the industry database, wherein the search log table contains product ID, session ID, behavior type, behavior duration, and transaction flag; The behavior type includes click behavior; The behavior duration represents the viewing time after the user clicks on the product; The transaction flag indicates whether the user completes the purchase after clicking on the product, which represents completion of purchase, and no represents non-completion of purchase; The candidate product set associated with the product keyword Ai is represented as {a1, a2,..., am}, wherein m represents the total number of candidate products; Extract the product ID an of any one candidate product in the candidate product set {a1, a2,..., am}, wherein n is a count index and its value range is 1 to m; For an, the following steps are performed: Step 1: from the search log table, query the number of unique session IDs of the product keyword Ai as the total display times S; Step 2: from the search log table, query the number of records of the product keyword Ai, the product ID an and the click behavior type as the click times C_click of the candidate product with the product ID an; Step 3: calculate the click rate CTR, the calculation formula is CTR=C_click / S, wherein if S=0, set CTR=0; Step 4: from the search log table, extract all records of the product keyword Ai, the product ID an and the click behavior type, obtain the behavior duration of all records and take the arithmetic mean as the viewing time T_view, wherein if there is no click record, set T_view=0; Step 5: from the search log table, query the number of records of the product keyword Ai, the product ID an, the click behavior type and the transaction flag as the transaction times C_purchase; Step 6: calculate the transaction rate CR, the calculation formula is CR=C_purchase / C_click, wherein if C_click=0, set CR=0; Repeat the above steps to determine the click rate, viewing time and transaction rate associated with each candidate product in the candidate product set under the search log of the corresponding product keyword; The specific way of constructing the search matching model to calculate the search matching index of each candidate product and the product keyword is: The click rate CTR, the view time T_view, and the transaction rate CR of any candidate product with a product ID an in the candidate product set {a1, a2,..., am} associated with the product keyword Ai; Calculate the maximum value T_max and the minimum value T_min of the view time of all candidate products in the candidate product set {a1, a2,..., am}, and use the maximum value T_max and the minimum value T_min to calculate the normalized view time T_norm associated with the view time T_view. The calculation steps are as follows: If T_max ≠ T_min, T_norm = (T_view - T_min) / (T_max - T_min); If T_max = T_min, T_norm = 0.5; The search matching index ML associated with the candidate product with a product ID an is calculated using ML = ω1 × CTR + ω2 × T_norm + ω3 × CR, where ω1, ω2, and ω3 are all preset weights, and ω1 + ω2 + ω3 = 1, ω1, ω2, and ω3 all belong to [0, 1]; Repeat the above steps to determine the search matching index of all candidate products in the candidate product set {a1, a2,..., am}; The search product sequence RPS is generated by combining the determined search matching index with the candidate product set and is displayed to the user; The specific way of extracting priority products from the search product sequence is as follows: The first product in all search product subsequences is extracted from the search product sequence RPS, and the priority product sequence a1', a2',..., aj' is arranged according to the arrangement order of the search product subsequence. The specific way of extracting a number of recommended products associated with the sequence number of the corresponding preferred product in the industry database based on the sequence number of the preferred product in the search product sequence is as follows: Obtain the priority product sequence a1', a2',..., aj'; Determine the weight coefficient of the recommended product associated with each of the j priority products; Where the weight coefficient Wi of the recommended product of any priority product with a product ID ai' is Wi = 1 / i; Calculate the sum W_total of the weight coefficients of the j priority products, W_total = Σ(Wi), i ∈ [1, j]; Calculate the recommended quantity proportion Pi of any priority product, Pi = Wi / W_total; Obtain the total number of recommended products N_total, and calculate the number of recommended products N_total_i of any priority product, N_total_i = floor(Pi * N_total), where floor is the floor function; For any priority product with a product ID ai', extract the top N_total_i products with the highest association weight from the product association table as its associated recommended products; Similarly, determine the recommended products associated with all priority products; And generate a recommended product sequence based on the sequence number of each priority product, and display it to the user synchronously with the search product sequence.

2. The method of claim 1, wherein, In step one, the specific way of extracting the product keyword in the user input content based on the interactive interface is as follows: An interactive interface for user interaction is acquired, and a content to be searched for input by the user is acquired based on the interactive interface; A set of lexical units is determined based on the content to be searched for, wherein the set of lexical units includes single-character units, two-character combinations, and complete noun phrases; A pre-constructed commodity core vocabulary table and a modifier filtering table in an industry database are acquired, wherein the commodity core vocabulary table contains industry-standard commodity names, and the modifier filtering table contains descriptive words; Descriptive words in the set of lexical units are filtered out using the modifier filtering table; All lexical units in the filtered set of lexical units are matched with the commodity core vocabulary table to determine all commodity words, and the commodity words are combined into a commodity word set {A1, A2,..., Aj} in the order of their appearance in the content to be searched for; All commodity words in the commodity word set are regarded as commodity keywords in the content to be searched for.

3. The method of claim 2, wherein, In step one, the specific method for determining all candidate commodities that meet the commodity keywords and generating a candidate commodity set is as follows: A commodity inventory table is acquired from an industry database; Any commodity keyword Ai in the commodity word set {A1, A2,..., Aj} is extracted, where i is a count index and takes a value ranging from 1 to j; A fuzzy search query is performed on the commodity keyword Ai in the commodity inventory table to extract all commodities that meet the commodity keyword Ai and have an inventory of no less than 1 as candidate commodities, and the product IDs of all candidate commodities are randomly combined to generate a candidate commodity set associated with the commodity keyword Ai, wherein the product ID is unique; Similarly, candidate commodity sets associated with all commodity keywords in the commodity word set {A1, A2,..., Aj} are determined.

4. The method of claim 3, wherein, In step two, the specific method for generating a search commodity sequence using the determined search matching index and the candidate commodity set is as follows: The candidate commodity set {a1, a2,..., am} associated with the first commodity keyword A1 in the commodity word set {A1, A2,..., Aj} is extracted; All candidate commodities in the candidate commodity set {a1, a2,..., am} are arranged in descending order of their respective search matching indexes, and the arranged result is recorded as a search commodity sub-sequence associated with the commodity keyword A1; Similarly, search commodity sub-sequences associated with all commodity keywords in the commodity word set {A1, A2,..., Aj} are determined, a total of j sub-sequences, each sub-sequence is used as a parameter in the search commodity sequence, and j sub-sequences are combined to generate a search commodity sequence RPS associated with the commodity word set {A1, A2,..., Aj}.

5. The method of claim 4, wherein, After determining the search commodity sequence RPS in step two, the search commodity sub-sequences in the search commodity sequence RPS are displayed to the user in a paginated form in the display part of the interactive interface in the order of their arrangement.

6. The method of claim 5, wherein, In step three, the specific method for generating a recommended commodity sequence is as follows: With the product ID of each priority product in the priority product sequence a1', a2',..., aj' as input, extract the recommended products associated with the serial number of each priority product from the product association table in the industry database, wherein the product association table records the association relationship between two products and the association weight, and the association weight is determined by the number of times of simultaneous purchase of two products by the user within a preset time period, the more the number of simultaneous purchase, the greater the association weight, and vice versa; Arrange all recommended products in the order of the priority product sequence a1', a2',..., aj', that is, arrange all recommended products associated with a1' first, then arrange all recommended products associated with a2', and so on, to generate a recommended product sequence; Among them, the arrangement order of the recommended products associated with the priority product with product ID a1' is in a random arrangement manner, and the rest is the same; Display the generated recommended product sequence to the user in the form of a pop-up window.

Images