Supply and demand content feature processing method and device based on enterprise dictionary system, computer equipment and readable storage medium

Abstract

The invention discloses a supply and demand content feature processing method and device based on an enterprise dictionary system, computer equipment and a readable storage medium, and the method comprises the steps: firstly obtaining the content of supply and demand data and title data through a DB link, and carrying out the splicing of the content and title data, and obtaining the initial splicing supply and demand content; and after target splicing supply and demand contents are obtained through filtering, tag and stop word dictionary data are generated based on the enterprise related dictionary. And performing word segmentation on target contents and filtering stop words by utilizing a word segmentation device in combination with the two types of dictionary data to obtain a word list. The method comprises the following steps: inputting a word list into an ailab embedding model to obtain a word vector matrix, respectively carrying out task and batch processing on the word vector matrix to obtain a full-amount three-dimensional matrix and a batch three-dimensional matrix, merging to obtain a target text vector, and finally storing the target text vector into a Redis DB database based on an item cache identifier to realize efficient and accurate feature processing of supply and demand contents.

Description

Technical Field

[0001] The present invention relates to the field of data processing, and in particular to a method, apparatus, computer equipment and readable storage medium for processing supply and demand content features based on an enterprise dictionary system. Background Art

[0002] As businesses grow, the volume of supply and demand data becomes increasingly large and complex. Traditional supply and demand data processing methods struggle to accurately extract key features from this massive and diverse data. Existing data processing technologies lack a systematic approach based on enterprise dictionaries, making it difficult to fully leverage information such as business tags and stop words to improve processing effectiveness. Consequently, they struggle to meet enterprises' needs for efficient and accurate processing of supply and demand data. Summary of the Invention

[0003] The object of the present invention is to provide a method, apparatus, computer equipment and readable storage medium for processing supply and demand content features based on an enterprise dictionary system.

[0004] In a first aspect, an embodiment of the present invention provides a method for processing supply and demand content features based on an enterprise dictionary system, comprising:

[0005] Obtaining content data and title data of the supply and demand data through a DB link, and splicing the content data and the title data to obtain initial spliced ​​supply and demand content; the content data and the title data constitute the original supply and demand business description content;

[0006] Filtering the initial splicing supply and demand data to obtain target splicing supply and demand content;

[0007] Generate tag dictionary data and stop word dictionary data based on enterprise-related dictionaries;

[0008] Based on the word segmenter combined with the tag dictionary data and the stop word dictionary data, the target spliced ​​supply and demand content is segmented and stop word filtered to obtain a word list;

[0009] Input the word list into the pre-trained ailab embedding model to obtain the word vector matrix corresponding to the word list;

[0010] Performing task processing on the word vector matrix to obtain a full three-dimensional matrix, and performing batch processing on the word vector matrix to obtain a batch three-dimensional matrix;

[0011] Merging the full three-dimensional matrix and the batch three-dimensional matrix to obtain a target text vector;

[0012] Based on the cache identifier of the item corresponding to the original supply and demand business description content, the target text vector is stored in a RedisDB database.

[0013] In a possible implementation, filtering the initial splicing supply and demand data to obtain target splicing supply and demand content includes:

[0014] By using a regular expression matching method, data consisting entirely of numbers, entirely of consecutive letters, entirely of consecutive letters and numbers, and special characters in the initial spliced ​​supply and demand data are identified and filtered out to obtain the target spliced ​​supply and demand content.

[0015] In a possible implementation, generating tag dictionary data and stop word dictionary data based on the enterprise-related dictionary includes:

[0016] Load the username dictionary to generate username dict, load the company name and company abbreviation dictionary to generate company dict, load the region and city dictionary to generate city dict, and load the industry dictionary to generate industry dict;

[0017] The user name dict, the company dict, the city dict, and the industry dict are used as the label dictionary data;

[0018] The stop word dictionary data is generated by merging the open source stop word library of the preset enterprise word segmenter.

[0019] In a possible implementation, when loading each dictionary, if the dictionary data is stored in a local file, the data is parsed and read according to the file format; if it is stored in a remote database, the data is extracted through the corresponding database interface.

[0020] In a possible implementation, before inputting the word list into a pre-trained AILABembedding model to obtain a word vector matrix corresponding to the word list, the method further includes:

[0021] The word list is deduplicated to reduce computational complexity and redundant data.

[0022] In a possible implementation, performing task processing on the word vector matrix to obtain a full three-dimensional matrix includes:

[0023] Input the word vector matrix as a whole into the task processing module, and generate the full three-dimensional matrix containing all word vector relationships at one time according to the set three-dimensional matrix generation algorithm;

[0024] The word vector matrix is ​​batch processed to obtain a batch three-dimensional matrix, including:

[0025] The word vector matrix is ​​divided into multiple batches of equal or similar size, and each batch is input into the batch processing module in turn to generate a corresponding three-dimensional matrix to obtain the batch three-dimensional matrix.

[0026] In a possible implementation, storing the target text vector in a RedisDB database based on a cache identifier of an item corresponding to the original supply and demand business description content includes:

[0027] Reading a cache identifier of an item corresponding to the original supply and demand business description from the cache, and if the item does not exist in the cache, querying the corresponding database table and storing it in the cache;

[0028] The cache identifier is used as a key and the target text vector is used as a value to store them in the RedisDB database.

[0029] In a second aspect, an embodiment of the present invention provides a device for processing supply and demand content features based on an enterprise dictionary system, comprising:

[0030] An acquisition module is configured to acquire content data and title data of supply and demand data through a DB link, and to concatenate the content data and title data to obtain initial concatenated supply and demand content; the content data and title data constitute original supply and demand business description content; filter the initial concatenated supply and demand data to obtain target concatenated supply and demand content; generate tag dictionary data and stop word dictionary data based on an enterprise-related dictionary; and perform word segmentation and stop word filtering on the target concatenated supply and demand content based on a word segmenter combined with the tag dictionary data and the stop word dictionary data to obtain a word list;

[0031] A processing module is used to input the word list into a pre-trained ailabedding model to obtain a word vector matrix corresponding to the word list; perform task processing on the word vector matrix to obtain a full three-dimensional matrix, and perform batch processing on the word vector matrix to obtain a batch three-dimensional matrix; merge the full three-dimensional matrix and the batch three-dimensional matrix to obtain a target text vector; based on the cache identifier of the item corresponding to the original supply and demand business description content, store the target text vector in a RedisDB database.

[0032] In a third aspect, an embodiment of the present invention provides a computer device, comprising a processor and a non-volatile memory storing computer instructions, wherein when the computer instructions are executed by the processor, the computer device executes the method described in the first aspect.

[0033] In a fourth aspect, an embodiment of the present invention provides a readable storage medium, wherein the readable storage medium includes a computer program, and when the computer program is executed, the computer device where the readable storage medium is located is controlled to execute the method described in the first aspect.

[0034] Compared with the prior art, the beneficial effects provided by the present invention include: using a method, device, computer equipment and readable storage medium for processing supply and demand content features based on an enterprise dictionary system disclosed in the present invention, obtaining the content and title data of the supply and demand data through a DB link and splicing them to obtain the initial spliced ​​supply and demand content. After filtering to obtain the target spliced ​​supply and demand content, generate label and stop word dictionary data based on the enterprise-related dictionary. Use a word segmenter combined with two types of dictionary data to segment the target content and filter stop words to obtain a word list. Input the word list into the ailab embedding model to obtain a word vector matrix, perform task and batch processing on it respectively to obtain full and batch three-dimensional matrices, merge them to obtain the target text vector, and finally store it in the RedisDB database based on the item cache identifier to achieve efficient and accurate feature processing of supply and demand content. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] To more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly describes the drawings required for use in the embodiments. It should be understood that the following drawings illustrate only certain embodiments of the present invention and should not be construed as limiting the scope of the present invention. Those skilled in the art can, without inventive effort, derive other relevant drawings from these drawings.

[0036] Figure 1 A schematic flow chart of the steps of a method for processing supply and demand content features based on an enterprise dictionary system provided by an embodiment of the present invention;

[0037] Figure 2 Schematic diagram of the supply and demand content embedding process based on enterprise tags and dictionaries provided in an embodiment of the present invention;

[0038] Figure 3 A schematic block diagram of the structure of an apparatus for processing supply and demand content features based on an enterprise dictionary system provided by an embodiment of the present invention;

[0039] Figure 4 A schematic block diagram of the structure of a computer device provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0040] To make the objectives, technical solutions, and advantages of the embodiments of the present invention more apparent, the technical solutions of the embodiments of the present invention will be described clearly and completely below in conjunction with the accompanying drawings of the embodiments of the present invention. It should be understood that the described embodiments are only a portion of the embodiments of the present invention, not all of them. Generally, the components of the embodiments of the present invention described and illustrated in the drawings herein may be arranged and designed in a variety of different configurations.

[0041] The specific embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0042] In order to solve the technical problems in the above background technology, Figure 1 This is a flow chart of a method for processing supply and demand content features based on an enterprise dictionary system provided by an embodiment of the present disclosure. The method for processing supply and demand content features based on an enterprise dictionary system is introduced in detail below.

[0043] Step S201, obtaining content data and title data of supply and demand data through a DB link, and splicing the content data and the title data to obtain initial spliced ​​supply and demand content; the content data and the title data constitute the original supply and demand business description content;

[0044] Step S202, filtering the initial splicing supply and demand data to obtain target splicing supply and demand content;

[0045] Step S203: Generate tag dictionary data and stop word dictionary data based on the enterprise-related dictionary;

[0046] Step S204, based on the word segmenter combined with the tag dictionary data and the stop word dictionary data, performing word segmentation and stop word filtering on the target spliced ​​supply and demand content to obtain a word list;

[0047] Step S205: input the word list into a pre-trained ailab embedding model to obtain a word vector matrix corresponding to the word list;

[0048] Step S206: performing task processing on the word vector matrix to obtain a full three-dimensional matrix, and performing batch processing on the word vector matrix to obtain a batch three-dimensional matrix;

[0049] Step S207: merging the full three-dimensional matrix and the batch three-dimensional matrix to obtain a target text vector;

[0050] Step S208: storing the target text vector in a RedisDB database based on a cache identifier of the item corresponding to the original supply and demand business description content.

[0051] In the embodiment of the present invention, for example, it is assumed that the server is responsible for processing the supply and demand data of a large e-commerce platform. On this platform, many merchants publish product supply information, and a large number of users also publish demand information.

[0052] The server establishes a stable connection with the database (DB) and begins to retrieve supply and demand data. For example, Merchant A posts supply information for a smartwatch. The title data is "High-quality smartwatch supply," and the content data details the watch's brand, model, functions, material, price range, and other information. For example, "This smartwatch is XX brand, model X1, has heart rate monitoring, sleep tracking, and other functions. It is made of titanium alloy and costs between 1,000 and 2,000 yuan."

[0053] After the server obtains this part of the title and content data, it splices them together. In this example, the initial spliced ​​supply and demand content becomes "Supply of high-quality smart watches. This smart watch is XX brand, model X1, with heart rate monitoring, sleep tracking and other functions, made of titanium alloy, and priced between 1,000 and 2,000 yuan." This splicing method enriches the information content of the data, allowing subsequent processing to be based on a more comprehensive business description. The same processing method is used for demand information. For example, user B publishes a demand with the title "Purchase of large-screen tablet computer" and the content is "Hope to buy a tablet computer with a screen size of more than 10 inches, a running memory of no less than 8GB, and a storage capacity of 256GB or more." After splicing, a more complete demand description is obtained.

[0054] After the server obtains the initial spliced ​​supply and demand content, it begins filtering. In real-world scenarios, some invalid data may exist. For example, when Merchant C publishes supply information, due to an error, the title may contain all numbers such as "123456", while the content may contain all consecutive letters such as "aaaaa", or all consecutive letters, numbers, and special characters such as "123abc!@#".

[0055] The server detects and filters these situations. For example, for merchant C, when the server detects the title "123456" and the content "aaaaa," it determines this as invalid data and filters it out, preventing it from entering the subsequent processing flow. This reduces the impact of invalid data on the overall processing process, conserves server resources, and improves the efficiency of subsequent embedding conversions. For example, when processing large amounts of supply and demand data, if this data is not filtered, it will occupy the server's computing resources, resulting in increased processing time. After filtering, the server can focus on processing valid supply and demand business descriptions.

[0056] The server loads data from relevant data sources to generate a username dictionary. For example, the user management system database of an e-commerce platform stores various user information. The server reads this data, using the username as the key and associating information such as the user ID and user type (such as regular user, merchant user, VIP user, etc.) as the value. For example, if the user "Zhang San" has the user ID "1001" and the user type "regular user," then the generated username dictionary will contain one record: the key is "Zhang San" and the value is a collection of related information such as "1001" and "regular user."

[0057] The server loads data related to company names and abbreviations from the company information database. For example, e-commerce platforms contain information about numerous merchants. For example, "Huawei Technologies Co., Ltd." is abbreviated as "Huawei." When generating a company name dictionary, the server uses "Huawei Technologies Co., Ltd." and "Huawei" as keys, with values ​​such as the associated company ID "2001," the company address "Shenzhen, Guangdong Province," and the business scope "R&D and manufacturing of communications equipment, smartphones, etc." This allows the server to quickly retrieve detailed company information when encountering supply and demand descriptions related to "Huawei."

[0058] The server extracts data from regional and city information sources to generate a city dictionary. For example, it retrieves data from a geographic information database, using the city name "Beijing" as the key and associating the region code "110000," the geographic coordinates "39°56′N, 116°20′E," and the regional characteristics "China's capital, political and cultural center" as the value. When "Beijing" is mentioned in a supply or demand description, the server can retrieve this relevant information from the city dictionary.

[0059] The server generates an industry dictionary based on industry-related data sources. For example, in the industry classification standard database, for "electronic information industry," the industry name is used as the key, and the industry code "IT01," industry standard terms such as "integrated circuits, software development," and industry development trends such as "development toward artificial intelligence and 5G applications" are associated as values. When supply and demand data involves content related to the electronic information industry, the server can use the industry dictionary to obtain relevant industry knowledge.

[0060] The user name dictionary, company name dictionary, city dictionary, and industry dictionary generated above together constitute the label dictionary data.

[0061] The server combines open-source stopword libraries such as those of Tencent, Sogou, and IK word segmenter to generate a stopword dictionary. For example, these open-source stopword libraries contain stopwords such as "的", "了", "在", "是", etc., which have no actual semantic meaning or have little impact on the semantic expression of the text. The server integrates these stopwords together to generate a stopword dictionary. In subsequent text processing, when these stopwords are encountered, filtering operations can be performed.

[0062] The server loads the stopword dictionary and the merged label dictionary into memory for quick access. Taking the intelligent watch supply information "High-quality intelligent watch supply. This intelligent watch is of XX brand, model X1, with functions such as heart rate monitoring and sleep tracking, made of titanium alloy material, and the price is between 1000 - 2000 yuan" previously released by merchant A as an example, the server uses the jieba word segmenter to segment it.

[0063] During the word segmenting process, in combination with the label dictionary data, if "Huawei" is mentioned in the supply and demand content, the word segmenter can recognize "Huawei" as a whole according to the company name dictionary, rather than splitting it into other meaningless parts, improving the accuracy of word segmenting. After word segmenting, a preliminary word list is obtained, such as "High-quality", "intelligent watch", "supply", "This", "intelligent watch", "is", "XX brand", "model", "X1", "has", "heart rate monitoring", "sleep tracking", "etc.", "functions", "uses", "titanium alloy", "material", "price", "is", "between", "1000", "-", "2000", "yuan", "".

[0064] Then, the server compares this word list with the stopword dictionary and filters out the stopwords. In this example, stopwords such as "This", "is", "etc.", "is" will be filtered out, and finally the processed word list "High-quality", "intelligent watch", "supply", "XX brand", "model", "X1", "has", "heart rate monitoring", "sleep tracking", "functions", "uses", "titanium alloy", "material", "price", "1000", "-", "2000", "yuan", "" is obtained.

[0065] The server loads the pre-trained AILAB embedding model into memory. Taking the processed word list "High-quality", "intelligent watch", "supply", "XX brand", "model", "X1", "has", "heart rate monitoring", "sleep tracking", "functions", "uses", "titanium alloy", "material", "price", "1000", "-", "2000", "yuan", "" as an example, the server inputs these words into the AILAB embedding model sequentially through the restful method.

[0066] Based on its trained algorithms and knowledge, the model generates a corresponding vector representation for each word. For example, the word "smartwatch" might be represented as a specific vector [0.1, 0.3, -0.2, 0.4, ...]. The value of each dimension represents a characteristic of the word in a low-dimensional semantic space. The vectors generated by all words are sequentially organized into a word vector matrix, with each row representing a word vector. This converts the textual word list into a numerical matrix that computers can better process.

[0067] The server performs full task processing on the entire word vector matrix. Assume that the word vector matrix has m rows (representing m words) and n columns (the dimension of each word vector). The server inputs the entire matrix as a whole into a specific processing module according to the set algorithm. For example, the algorithm may consider factors such as the spatial relationship and semantic association between word vectors to generate a full three-dimensional matrix at one time. This three-dimensional matrix can reflect the comprehensive relationship between word vectors from a global perspective. For example, it may contain the association information of different word vectors at different semantic levels. Each element of the matrix represents a numerical feature of a specific word vector relationship.

[0068] Due to limited server resources, direct full processing of large word vector matrices may cause performance issues. Therefore, the server performs batch processing on the word vector matrix. The server divides the word vector matrix into multiple small data batches (batches), each batch containing several rows of word vectors. For example, the word vector matrix is ​​divided into k batches, each batch containing p rows of word vectors (m=k*p).

[0069] The server processes each batch in turn, generating a corresponding three-dimensional matrix for each batch according to a specific algorithm. This process is similar to full-scale processing, but only operates on the word vectors of each small batch. In this way, batch processing can efficiently process large amounts of data with limited resources, generating multiple small three-dimensional matrices that reflect the relationships between word vectors from a local perspective.

[0070] The server merges the full three-dimensional matrix generated by full processing and the batch three-dimensional matrix generated by batch processing. For example, the elements of the corresponding positions of the two matrices are fused by adding the matrix elements. Assume that the full three-dimensional matrix is ​​A, the batch three-dimensional matrix is ​​B, and the generated target text vector matrix is ​​C. For each position (i, j, k) in the matrix, C(i, j, k) = A(i, j, k) + B(i, j, k). Through this merging method, both the global word vector relationship (full three-dimensional matrix) and the local word vector relationship (batch three-dimensional matrix) are taken into account, and finally a comprehensive target text vector is obtained, which comprehensively reflects the semantic characteristics of the original supply and demand data content.

[0071] In the e-commerce platform scenario, each supply and demand information has a corresponding item identifier, such as the product ID for supply information and the order number for demand information. Taking the smartwatch supply information published by Merchant A as an example, assume that its corresponding item cache identifier is "item12345".

[0072] The server stores the generated target text vector in the RedisDB database, using "item12345" as the key and the target text vector as the value. This allows the server to quickly query and retrieve the corresponding text vector from RedisDB based on the item's cached identifier when performing subsequent business operations such as supply and demand matching or data analysis, providing data support for these business functions. For example, when performing supply and demand matching, the server can quickly retrieve text vectors for different supply and demand information and identify potential matching pairs by calculating similarities between the vectors.

[0073] In a possible implementation, filtering the initial splicing supply and demand data to obtain target splicing supply and demand content may be performed through the following example.

[0074] By using a regular expression matching method, data consisting entirely of numbers, entirely of consecutive letters, entirely of consecutive letters and numbers, and special characters in the initial spliced ​​supply and demand data are identified and filtered out to obtain the target spliced ​​supply and demand content.

[0075] In an embodiment of the present invention, for example, assume that a server is processing supply and demand information for a comprehensive commercial service platform. The platform contains both supply information for various commodities and demand postings for enterprise services. For example, an electronics supplier posts information with the initial supply and demand content being "Supplying a brand new smartphone, model: XYZ-10, with 5G functionality, price: 3,000 yuan, contact number: 138." However, some invalid information may also exist. For example, data consisting entirely of numbers, such as "1234567890," may be a string of numbers entered by the user by mistake and not a valid supply and demand business description. There are also cases where data consisting entirely of consecutive letters, such as "abcdefghij," may be the result of a user input error. Furthermore, data consisting entirely of consecutive letters, numbers, and special characters, such as "abc123!@#," may also appear, which is also meaningless to the supply and demand business.

[0076] The server uses regular expression matching to handle these situations. For data that is entirely numeric, the server uses the regular expression "^\d+$" to match. When data such as "1234567890" is detected, the regular expression will recognize that it meets the pattern, and the server will filter it out. For data that is entirely consecutive letters, the regular expression "^[a-zA-Z]+$" is used. If "abcdefghij" is encountered, this expression can identify it and allow the server to remove it from the initial spliced ​​supply and demand data. For data that is entirely consecutive letters, numbers, and special characters, the server uses more complex regular expressions, such as "^[a-zA-Z0-9!@#$%^&*()_+-=]+$". When "abc123!@#" appears, the regular expression matches successfully, and the server will filter out such invalid data.

[0077] Through this series of regular expression-based filtering operations, the server removes meaningless and invalid data from the initial spliced ​​supply and demand data, resulting in the target spliced ​​supply and demand content. This target spliced ​​supply and demand content retains only information that is truly valuable for describing the supply and demand business. For example, the filtered supply information of the aforementioned electronics supplier can be used in subsequent more accurate and efficient data processing processes, laying a solid foundation for the processing of supply and demand content features within the enterprise dictionary system.

[0078] In a possible implementation, the generation of tag dictionary data and stop word dictionary data based on the enterprise-related dictionary may be implemented through the following examples.

[0079] Load the username dictionary to generate username dict, load the company name and company abbreviation dictionary to generate company dict, load the region and city dictionary to generate city dict, and load the industry dictionary to generate industry dict;

[0080] The user name dict, the company dict, the city dict, and the industry dict are used as the label dictionary data;

[0081] The stop word dictionary data is generated by merging the open source stop word library of the preset enterprise word segmenter.

[0082] In an embodiment of the present invention, for example, in a large-scale commercial data processing scenario, a server is responsible for processing supply and demand information from numerous enterprises in various industries. First, the server loads different types of dictionaries respectively.

[0083] The server loads data from the user information database to generate a username dictionary. For example, the database records the information of numerous platform users. For instance, for "Li Ming", the corresponding user ID is "U001" and the user type is "individual merchant". The server uses "Li Ming" as the key and related information such as "U001" and "individual merchant" as the values to construct the username dictionary.

[0084] Next, the server loads data related to company names and abbreviations from the company information database to generate a company dictionary. For example, there is "Alibaba Group Holding Limited", with the abbreviation "Alibaba", the company ID is "C001", and its business covers e-commerce, fintech and other fields. The server uses "Alibaba Group Holding Limited" and "Alibaba" as keys and associates information such as "C001" and business fields to form the company dictionary.

[0085] The server loads data on regions and cities from the geographical information data source to generate a city dictionary. For example, for "Shanghai", the region code is "310000", the geographical location coordinates are "31°14′ north latitude, 121°29′ east longitude", and the regional characteristics are "international metropolis, economic and financial center". The server uses "Shanghai" as the key and includes information such as region code, coordinates and regional characteristics as the values to generate the city dictionary.

[0086] The server loads the industry dictionary based on the industry classification data to generate an industry dictionary. For example, for the "Internet industry", the industry code is "IT002", the standard terms are "cloud computing, big data", and the development trend is "transforming towards industrial Internet". The server uses "Internet industry" as the key and associates the industry code, standard terms and development trend, etc. to form the industry dictionary.

[0087] After that, the server integrates the username dictionary, company dictionary, city dictionary and industry dictionary together as the label dictionary data. These data provide multi-dimensional reference information for subsequent text processing. For example, when analyzing supply and demand texts, the company dictionary can be used to accurately identify company names and improve the accuracy of word segmentation.

[0088] Finally, the server combines the stop word libraries open-sourced by the preset enterprise word segmenters to generate stop word dictionary data. For example, in the stop word libraries open-sourced by Tencent, Sogou and IK word segmenters, there are words such as "de", "le", "zai", "ye", etc. The server integrates these stop words together to construct the stop word dictionary data. When processing supply and demand texts subsequently, using this stop word dictionary data can filter out words without actual semantic contributions, purify the text and improve the processing effect.

[0089] In a possible implementation, when loading each dictionary, if the dictionary data is stored in a local file, the data is parsed and read according to the file format; if it is stored in a remote database, the data is extracted through the corresponding database interface.

[0090] In an embodiment of the present invention, illustratively, when the server processes supply and demand data and generates tag dictionary data and stop word dictionary data, different methods need to be adopted according to the data storage location when loading various dictionaries.

[0091] Suppose the server serves a large enterprise service platform that integrates a vast amount of enterprise supply and demand information. Username dictionary data, if stored in a local file, might be stored in CSV format. Each row contains information such as the username, user ID, and user type, separated by commas. The server parses and reads the data according to the CSV file format, reading it line by line, using the username as the key and the user ID, user type, and other information as the value to generate a username dict. For example, if the file contains a row of data "Zhang San, 1001, Enterprise User", the server reads it and associates "Zhang San" with "1001" and "Enterprise User" in the username dict.

[0092] If the company name and company abbreviation dictionary data is stored in a remote relational database (such as MySQL), the server retrieves the data through the corresponding MySQL database interface. The server first establishes a connection with the database and uses the SQL query "SELECT company_name, company_short_name, company_id, business_scope FROM company_dict_table" to extract the company name, abbreviation, ID, and business scope from the table named "company_dict_table." After extraction, the company dict is generated using the company name and abbreviation as keys and the company ID and business scope as values. For example, if the query results include "Huawei Technologies Co., Ltd., Huawei, 2001, Communications Equipment, Smartphone R&D and Manufacturing," the server constructs the relevant records in the company dict based on this information.

[0093] If region and city dictionary data is stored locally in JSON file format, the server parses and reads it according to the JSON file format. The file structure may be an object containing information about multiple cities, with each city object containing attributes such as the city name, region code, geographic coordinates, and regional characteristics. After reading the file, the server generates a city dict using the city name as the key and an object consisting of other attribute values ​​as the value. For example, if the file contains "{'city name': 'Beijing', 'region code': '110000', 'geographic coordinates': '39°56′N, 116°20′E', 'regional characteristics': 'China's capital, political and cultural center'}," the server uses this to generate records related to "Beijing" in the city dict.

[0094] For industry dictionary data, if it's stored in a remote, non-relational database (such as MongoDB), the server extracts the data through the MongoDB database interface. The server connects to the MongoDB database and uses the corresponding query method to retrieve data such as industry names, industry codes, standard terms, and development trends from the "industry_dict_collection" collection. After extraction, an industry dict is generated using the industry name as the key and other information as the value. For example, if "{'Industry Name':'Electronic Information Industry', 'Industry Code':'IT01', 'Standard Terms':'Integrated Circuits, Software Development', 'Development Trends':'Towards Artificial Intelligence and 5G Applications'}" is obtained, the server constructs the records in the industry dict based on this. Through these methods, the server successfully loads various dictionary data, laying the foundation for subsequent processing.

[0095] In an embodiment of the present invention, before inputting the word list into the pre-trained AILABembedding model to obtain the word vector matrix corresponding to the word list, the following implementation method is also provided.

[0096] The word list is deduplicated to reduce computational complexity and redundant data.

[0097] In an embodiment of the present invention, for example, in the process of processing supply and demand data on the server, after obtaining a word list that has been segmented and stop word filtered, an important step is to deduplicate the word list before inputting it into a pre-trained AILABembedding model.

[0098] For example, consider the server of a comprehensive e-commerce platform. Merchants and users on the platform publish a rich and diverse supply and demand information. For example, a merchant posts supply information for a multifunction printer. After word segmentation and stop word filtering, the resulting word list is: "multifunction," "printer," "supply," "print," "copy," "scan," "function," "black and white," "color," "print," "high speed," and "print."

[0099] The server starts to deduplicate the word list. It goes through the entire word list and checks each word one by one to see if it is already in a temporary storage structure (such as a collection). Starting with the first word "multi-function", since the collection is empty, "multi-function" is added to the collection. Then it processes "printer", and since there is no such word in the collection, "printer" is also added to the collection. When it processes the first "print", there is no such word in the collection, so "print" is added. But when "print" is encountered again, the server finds that "print" already exists in the collection, so it removes it from the word list. And so on, the same processing is performed for repeated "print".

[0100] After deduplication, the word list becomes: "multi-function," "printer," "supply," "print," "copy," "scan," "function," "black and white," "color," and "high speed." This significantly shortens the word list and reduces a lot of redundant data.

[0101] This deduplication process has a significant positive impact on the subsequent input of the word list into the AILABembedding model. When the AILABembedding model generates the word vector matrix, each word requires complex calculation operations. If there are a large number of repeated words in the word list, the model will perform the same calculations on these repeated words multiple times, which not only wastes the server's computing resources but also increases processing time. Through deduplication, the model only needs to calculate non-repeated words, greatly reducing the amount of calculation. This optimization effect is even more obvious when processing massive amounts of supply and demand data. It can significantly improve the processing efficiency of the server, making the entire supply and demand content feature processing process more efficient and smooth, and laying a good foundation for the subsequent generation of accurate word vector matrices and further data processing.

[0102] In a possible implementation, the task processing is performed on the word vector matrix to obtain a full three-dimensional matrix, which can be implemented through the following example.

[0103] Input the word vector matrix as a whole into the task processing module, and generate the full three-dimensional matrix containing all word vector relationships at one time according to the set three-dimensional matrix generation algorithm;

[0104] The word vector matrix is ​​batch processed to obtain a batch three-dimensional matrix, including:

[0105] The word vector matrix is ​​divided into multiple batches of equal or similar size, and each batch is input into the batch processing module in turn to generate a corresponding three-dimensional matrix to obtain the batch three-dimensional matrix.

[0106] In this embodiment of the present invention, for example, assume that the server is processing supply and demand data from a large manufacturing enterprise. After the previous process, a word vector matrix describing the enterprise's product supply and service demand has been obtained.

[0107] To generate the full three-dimensional matrix, the server inputs the word vector matrix as a whole into the task processing module. For example, this word vector matrix includes various word vectors describing the high-end CNC machine tools supplied by the company, such as vectors corresponding to terms like "high precision," "automation," "milling," and "turning." The task processing module operates according to a predefined three-dimensional matrix generation algorithm. This algorithm may construct the three-dimensional matrix based on factors such as the positional relationships between word vectors in semantic space and the co-occurrence frequency between words. Taking the word vectors of "high precision" and "automation" as examples, the algorithm analyzes the closeness of their descriptions of the performance characteristics of CNC machine tools and numerically represents this relationship at the corresponding positions in the three-dimensional matrix. Ultimately, a full three-dimensional matrix containing all word vector relationships is generated at once. This matrix comprehensively reflects the complex semantic connections between word vectors and comprehensively covers the various word relationships in the description of the supply and demand of high-end CNC machine tools.

[0108] Next, batch processing is performed. Since the word vector matrix may be large in scale, direct processing will put a lot of pressure on the server. The server divides the word vector matrix into multiple batches of equal or similar size. For example, the word vector matrix describing high-end CNC machine tools is evenly divided into 5 batches according to the number of vocabulary, and each batch contains a part of the word vectors. Then, the server inputs each batch into the batch processing module in turn. Taking the first batch as an example, it may contain word vectors such as "high precision" and "milling". For this part of the word vectors, the batch processing module generates a corresponding three-dimensional matrix according to the algorithm logic similar to the full processing, but only within the word vector range of this batch. This three-dimensional matrix reflects the relationship between the word vectors in the batch. In this way, the server processes each batch and finally obtains multiple three-dimensional matrices, which together constitute the batch three-dimensional matrix. Through this batch processing method, the server can process large-scale word vector matrices more efficiently under limited resources. The three-dimensional matrix generated by each batch can also reflect the relationship between word vectors locally, complementing the full three-dimensional matrix, and providing strong support for subsequent merging to generate target text vectors that fully reflect the semantic characteristics of supply and demand data.

[0109] In the embodiment of the present invention, storing the target text vector in the RedisDB database based on the cache identifier of the item corresponding to the original supply and demand business description content can be implemented through the following example.

[0110] Reading a cache identifier of an item corresponding to the original supply and demand business description from the cache, and if the item does not exist in the cache, querying the corresponding database table and storing it in the cache;

[0111] The cache identifier is used as a key and the target text vector is used as a value to store them in the RedisDB database.

[0112] In the embodiment of the present invention, for example, it is assumed that the server is processing the supply and demand data of a large e-commerce platform, where many merchants publish commodity supply information, and each supply information has a corresponding item.

[0113] After processing the supply and demand data to obtain the target text vector, the server begins to perform storage operations. For example, the supply information of a smart speaker released by Merchant A is an original supply and demand business description.

[0114] First, the server attempts to retrieve the cached identifier for the item corresponding to the smart speaker's provisioning information from its cache. The cache acts as a "fast memory" for the server, speeding up data access. If the identifier, such as "item_speaker_123," exists in the cache, the server can retrieve it directly.

[0115] However, if the identifier does not exist in the cache, the server will query the corresponding database table. Suppose there is a table called "supply_items" in the database, which stores detailed information about all product supplies, including the item's unique identifier. The server uses an SQL query statement such as "SELECTitem_id FROM supply_items WHERE supply_description LIKE'% smart speaker%'" (this example is simplified here, and more precise matching conditions may be used in practice) to find the item identifier "item_speaker_123" corresponding to the smart speaker supply information from the table, and then stores it in the cache for quick reading next time.

[0116] Next, the server uses the cache identifier "item_speaker_123" obtained as the key and the target text vector of the supply information of this smart speaker obtained after a series of processing as the value, and stores it in the RedisDB database. The RedisDB database is very suitable for storing this type of data due to its fast data reading and writing performance. In this way, when the subsequent platform needs to perform supply and demand matching, data analysis and other operations, it can quickly obtain the corresponding target text vector from the RedisDB database based on this cache identifier. For example, when performing supply and demand matching, the system can quickly find the target text vectors of different product supply information, and by calculating the similarity between the vectors, find the supply information that may match the demand for smart speakers, providing strong data support for the efficient operation of the e-commerce platform.

[0117] In order to more clearly describe the solution provided by the embodiment of the present invention, a more complete implementation method is provided below. Figure 2 , Figure 2 A schematic diagram of the supply and demand content embedding process based on enterprise tags and dictionaries provided in an embodiment of the present invention.

[0118] 1. Data Input

[0119] The supply and demand data content serves as the starting input for the entire processing flow. This data content covers various types of information related to supply and demand, and may include specific text content such as the company's supply product description, demand service description, and other related ancillary information.

[0120] 2. Enterprise Dictionary Processing

[0121] 1. Username dictionary generation: Process the relevant data sources, load and generate a username dictionary (unamedict). This dictionary stores information related to the username, such as the username as the key and the associated user ID, user type, etc. as the value.

[0122] 2. Company Name Dictionary Generation: Using a specific data processing method, the company name and company abbreviation data are loaded to generate a company name dictionary (company dict). The company name or abbreviation is used as the key, and the corresponding information such as the associated company ID, company address, and business scope is used as the value.

[0123] 3. City dictionary generation: Extract region and city data from the corresponding data resources to generate a city dictionary (city dict). The dictionary uses the city or region name as the key and may associate region codes, geographic coordinates, regional characteristics, and other information as values.

[0124] 4. Industry dictionary generation: Generate an industry dictionary (industry dict) based on industry-related data sources. Use the industry name as the key and associate information such as industry codes, industry standard terms, and industry development trends as values.

[0125] The user name dictionary, company name dictionary, city dictionary, and industry dictionary generated above together constitute the enterprise dictionary system, providing basic reference information for subsequent data processing.

[0126] III. Data Processing

[0127] 1. Data flow: After the supply and demand data content undergoes preliminary data processing, it is transmitted in the form of a data stream to the analysis module based on the jieba tokenizer.

[0128] 2. Jieba tokenization analysis: Use the jieba tokenizer to tokenize the input supply and demand data content, splitting the continuous text content into individual words to form a keyword list (keyword).

[0129] 3. Stop word filtering: Compare the keyword list generated above with the stop word dictionary (stopword dict). The stop word dictionary is generated by merging open-source stop word libraries such as Tencent, Sogou, and IK tokenizer, which contains stop words such as "de", "le", "zai", etc. that have no actual semantics or have little impact on text semantic expression. By comparison, filter out the stop words in the keyword list to further purify the text data and obtain the processed keyword set.

[0130] IV. Model Processing and Vector Generation

[0131] 1. Model loading: Load the pre-trained ailab model (ailab model) into the system memory to provide model support for subsequent vectorization operations.

[0132] 2. Restful processing and vector generation: Process the keyword set after stop word filtering in a restful manner and generate corresponding word vectors with the help of the ailab model. These word vectors are organized in a specific structure to form a word vector matrix, where each vector represents the numerical representation of the corresponding word in the low-dimensional semantic space.

[0133] V. Matrix Processing and Merging

[0134] 1. Batch processing: Perform batch processing on the generated word vector matrix. Divide the word vector matrix into multiple small data batches (batch) and process each batch in turn to generate the corresponding three-dimensional matrix. This processing method helps to efficiently process large-scale data with limited resources.

[0135] 2. Full processing: At the same time, the word vector matrix is ​​fully processed. The entire word vector matrix is ​​input into a specific processing module as a whole. According to the set algorithm, a full three-dimensional matrix containing all word vector relationships is generated at once, reflecting the comprehensive relationship between word vectors from a global perspective.

[0136] 3. Matrix merging: Merge the three-dimensional matrix generated by batch processing with the full three-dimensional matrix generated by full processing. Using a specific merging algorithm, such as matrix element addition or weighted addition, the corresponding elements of the two matrices are combined to form a comprehensive text vector that fully reflects the semantic characteristics of the original supply and demand data content.

[0137] 6. Data Storage

[0138] The resulting text vector is stored in a RedisDB database. During storage, the item identifier associated with the original supply and demand data (e.g., the item ID) is used as the key, and the text vector is stored as the value. This facilitates subsequent rapid query and retrieval of the corresponding text vector based on the item identifier, providing data support for business functions such as supply and demand matching and data analysis.

[0139] In addition, another complete implementation is provided:

[0140] 1) Loading supply and demand data content:

[0141] Load content data and title data through db link;

[0142] The above loading logic is used to realize the splicing of content and title, enriching the supply and demand content.

[0143] 2) Basic data processing:

[0144] The supply and demand content filled in by the user is filtered, and the data consisting entirely of numbers, entirely of consecutive letters, entirely of consecutive letters and numbers, and special characters are filtered.

[0145] 3) Dictionary data preprocessing:

[0146] Load the user dictionary to generate a dict; load the company name and company abbreviation to generate a dict; load the industry dictionary to generate a dict; load the region and city dictionary to generate a dict; merge the preset open source stop word library of the enterprise word segmenter to generate a dict for the stop word dictionary;

[0147] 4) Perform word segmentation and stop word processing based on the jiaba word segmenter:

[0148] Load stop words and the merged dictionary into memory;

[0149] Filter the word list;

[0150] 5) Model loading and task part:

[0151] Load the ailabedding model into memory;

[0152] Generate embedding vectors by processing words in a restful way;

[0153] The embedding matrix generated by the word list;

[0154] 6) Generate full word vectors and batch vectors:

[0155] Generate 3D matrices in batches through all tasks;

[0156] Process small data sets of 3D matrices through batch processing;

[0157] Vectors are merged into text vectors;

[0158] 7) Store the supply and demand feature vectors in RedisDB:

[0159] The item ID is cached in redis.

[0160] In summary, the embodiments of the present invention demonstrate various technical effects and advantages when processing supply and demand data. In the supply and demand data content loading link, by splicing the content of the title and content, the information content of the content is greatly enriched, providing a more comprehensive data foundation for subsequent processing. In the basic data processing step, the data is filtered to reduce the interference of invalid data, which not only saves resources, but also significantly improves the embedding conversion efficiency. In the dictionary data preprocessing stage, the word segmentation accuracy is effectively increased through the careful processing of the label dictionary and the stop word dictionary. In terms of matrix feature fusion, the vectors are merged by summing n*100 dimensional vectors, which retains the text information to the greatest extent and fully prepares the data for the subsequent use of text features. The storage operation of the supply and demand feature vector can provide a fast storage effect for the use of data and incremental update data.

[0161] The key technical points and intended protections of this invention are primarily reflected in two aspects. First, a word segmentation and stop word processing scheme for supply and demand content data based on the company's vertical business dictionary improves the accuracy and pertinence of data processing. Second, information extraction using AILAB's word2vector model for supply and demand content demonstrates optimal information extraction results, enabling precise mining of key information within supply and demand data.

[0162] Please refer to Figure 3 , Figure 3 An embodiment of the present invention provides a device 110 for processing supply and demand content features based on an enterprise dictionary system, including:

[0163] Acquisition module 1101 is configured to acquire content data and title data of supply and demand data through a DB link, and to concatenate the content data and title data to obtain initial concatenated supply and demand content; the content data and title data constitute original supply and demand business description content; filter the initial concatenated supply and demand data to obtain target concatenated supply and demand content; generate tag dictionary data and stop word dictionary data based on an enterprise-related dictionary; and perform word segmentation and stop word filtering on the target concatenated supply and demand content based on a word segmenter combined with the tag dictionary data and the stop word dictionary data to obtain a word list;

[0164] Processing module 1102 is used to input the word list into a pre-trained ailab embedding model to obtain a word vector matrix corresponding to the word list; perform task processing on the word vector matrix to obtain a full three-dimensional matrix, and perform batch processing on the word vector matrix to obtain a batch three-dimensional matrix; merge the full three-dimensional matrix and the batch three-dimensional matrix to obtain a target text vector; based on the cache identifier of the item corresponding to the original supply and demand business description content, store the target text vector in the RedisDB database.

[0165] It should be noted that the implementation principles of the aforementioned device 110 for processing supply and demand content features based on an enterprise dictionary system can be referenced to the implementation principles of the aforementioned method for processing supply and demand content features based on an enterprise dictionary system, and will not be further elaborated here. It should be understood that the division of the various modules of the aforementioned device is merely a division of logical functions. In actual implementation, they can be fully or partially integrated into a single physical entity, or physically separated. Furthermore, these modules can be implemented entirely as software invoked by a processing element, or entirely as hardware, or partially as software invoked by a processing element, while others are implemented in hardware. For example, the device 110 for processing supply and demand content features based on an enterprise dictionary system can be a separate processing element, or integrated into a chip of the aforementioned device. Furthermore, it can be stored in the form of program code in the memory of the aforementioned device, with a processing element of the aforementioned device invoking and executing the functions of the aforementioned device 110 for processing supply and demand content features based on an enterprise dictionary system. The implementation of the other modules is similar. Furthermore, these modules can be fully or partially integrated or implemented independently. The processing element described here can be an integrated circuit with signal processing capabilities. During implementation, each step of the above method or each module can be completed by an integrated logic circuit of hardware in a processor element or by instructions in software form.

[0166] For example, the above modules may be one or more integrated circuits configured to implement the above methods, such as one or more application specific integrated circuits (ASICs), one or more digital signal processors (DSPs), or one or more field programmable gate arrays (FPGAs). For another example, when a module is implemented by scheduling program code on a processing element, the processing element may be a general-purpose processor, such as a central processing unit (CPU) or other processor that can call program code. For another example, these modules may be integrated together and implemented in the form of a system-on-a-chip (SOC).

[0167] The embodiment of the present invention provides a computer device 100, which includes a processor and a non-volatile memory storing computer instructions. When the computer instructions are executed by the processor, the computer device 100 executes the aforementioned device 110 for processing supply and demand content features based on the enterprise dictionary system. Figure 4 As shown, Figure 4This is a block diagram of a computer device 100 according to an embodiment of the present invention. The computer device 100 includes an apparatus 110 for processing supply and demand content features based on an enterprise dictionary system, a memory 111 , a processor 112 , and a communication unit 113 .

[0168] In order to realize the transmission or interaction of data, the memory 111, the processor 112 and the communication unit 113 are electrically connected to each other directly or indirectly. For example, the electrical connection between these components can be realized through one or more communication buses or signal lines. The device 110 for processing supply and demand content characteristics based on the enterprise dictionary system includes at least one software function module that can be stored in the memory 111 in the form of software or firmware or solidified in the operating system (OS) of the computer device 100. The processor 112 is used to execute the device 110 for processing supply and demand content characteristics based on the enterprise dictionary system stored in the memory 111, such as the software function modules and computer programs included in the device 110 for processing supply and demand content characteristics based on the enterprise dictionary system.

[0169] An embodiment of the present invention provides a readable storage medium, which includes a computer program. When the computer program is executed, the computer device where the readable storage medium is located is controlled to execute the aforementioned device 110 for processing supply and demand content features based on the enterprise dictionary system.

[0170] For illustrative purposes, the foregoing description has been made with reference to specific embodiments. However, the above illustrative discussion is not intended to be exhaustive or to limit the present disclosure to the precise forms disclosed. Numerous modifications and variations are possible in light of the above teachings. These embodiments have been selected and described in order to best illustrate the principles of the present disclosure and its practical application, thereby enabling those skilled in the art to best utilize the present disclosure and to utilize various embodiments with various modifications as appropriate for the specific application contemplated.

Claims

1. A method for processing supply and demand content features based on an enterprise dictionary system, characterized in that: include: Obtaining content data and title data of the supply and demand data through a DB link, and splicing the content data and the title data to obtain initial spliced ​​supply and demand content; the content data and the title data constitute the original supply and demand business description content; Filtering the initial splicing supply and demand data to obtain target splicing supply and demand content; Generate tag dictionary data and stop word dictionary data based on enterprise-related dictionaries; Based on the word segmenter combined with the tag dictionary data and the stop word dictionary data, the target spliced ​​supply and demand content is segmented and stop word filtered to obtain a word list; Input the word list into the pre-trained ailab embedding model to obtain the word vector matrix corresponding to the word list; Performing task processing on the word vector matrix to obtain a full three-dimensional matrix, and performing batch processing on the word vector matrix to obtain a batch three-dimensional matrix; Merging the full three-dimensional matrix and the batch three-dimensional matrix to obtain a target text vector; Based on the cache identifier of the item corresponding to the original supply and demand business description content, the target text vector is stored in a RedisDB database.

2. The method according to claim 1, characterized in that The filtering of the initial splicing supply and demand data to obtain target splicing supply and demand content includes: By using a regular expression matching method, data consisting entirely of numbers, entirely of consecutive letters, entirely of consecutive letters and numbers, and special characters in the initial spliced ​​supply and demand data are identified and filtered out to obtain the target spliced ​​supply and demand content.

3. The method according to claim 1, characterized in that The step of generating tag dictionary data and stop word dictionary data based on the enterprise-related dictionary includes: Load the username dictionary to generate username dict, load the company name and company abbreviation dictionary to generate company dict, load the region and city dictionary to generate city dict, and load the industry dictionary to generate industry dict; The user name dict, the company dict, the city dict, and the industry dict are used as the label dictionary data; The stop word dictionary data is generated by merging the open source stop word library of the preset enterprise word segmenter.

4. The method according to claim 3, characterized in that When loading each dictionary, if the dictionary data is stored in a local file, the data is parsed and read according to the file format; if it is stored in a remote database, the data is extracted through the corresponding database interface.

5. The method according to claim 1, wherein Before inputting the word list into the pre-trained AILABembedding model to obtain the word vector matrix corresponding to the word list, the method further includes: The word list is deduplicated to reduce computational complexity and redundant data.

6. The method according to claim 1, characterized in that The task processing is performed on the word vector matrix to obtain a full three-dimensional matrix, including: Input the word vector matrix as a whole into the task processing module, and generate the full three-dimensional matrix containing all word vector relationships at one time according to the set three-dimensional matrix generation algorithm; The word vector matrix is ​​batch processed to obtain a batch three-dimensional matrix, including: The word vector matrix is ​​divided into multiple batches of equal or similar size, and each batch is input into the batch processing module in turn to generate a corresponding three-dimensional matrix to obtain the batch three-dimensional matrix.

7. The method according to claim 1, characterized in that The step of storing the target text vector in a RedisDB database based on a cache identifier of an item corresponding to the original supply and demand business description content includes: Reading a cache identifier of an item corresponding to the original supply and demand business description from the cache, and if the item does not exist in the cache, querying the corresponding database table and storing it in the cache; The cache identifier is used as a key and the target text vector is used as a value to store them in the RedisDB database.

8. A device for processing supply and demand content features based on an enterprise dictionary system, characterized in that: include: An acquisition module is configured to acquire content data and title data of supply and demand data through a DB link, and to splice the content data and title data to obtain initial spliced ​​supply and demand content; the content data and title data constitute original supply and demand business description content; filter the initial spliced ​​supply and demand data to obtain target spliced ​​supply and demand content; and generate tag dictionary data and stop word dictionary data based on enterprise-related dictionaries; Based on the word segmenter combined with the tag dictionary data and the stop word dictionary data, the target spliced ​​supply and demand content is segmented and stop word filtered to obtain a word list; A processing module is used to input the word list into a pre-trained ailabedding model to obtain a word vector matrix corresponding to the word list; perform task processing on the word vector matrix to obtain a full three-dimensional matrix, and perform batch processing on the word vector matrix to obtain a batch three-dimensional matrix; merge the full three-dimensional matrix and the batch three-dimensional matrix to obtain a target text vector; based on the cache identifier of the item corresponding to the original supply and demand business description content, store the target text vector in a RedisDB database.

9. A computer device, characterized in that: The computer device includes a processor and a non-volatile memory storing computer instructions. When the computer instructions are executed by the processor, the computer device executes the method according to any one of claims 1 to 7.

10. A readable storage medium, characterized in that: The readable storage medium includes a computer program, and when the computer program is executed, the computer device where the readable storage medium is located is controlled to execute the method according to any one of claims 1 to 7.

Images