A method and system for mining structured information of goods in a second-hand e-commerce scenario

Abstract

The application relates to a second-hand e-commerce scene-based commodity structured information mining method and system, the method comprising: acquiring commodity information; in response to the fact that the commodity information comprises text content, performing information extraction on the text content to obtain a plurality of attributes and corresponding attribute information; in response to the fact that the commodity information comprises a commodity image, obtaining the commodity's item word by using an image classification model; replacing non-standardized words in the attribute information with standardized words synonymous therewith; and combining the item word and the attribute information of the current commodity to form structured information of the commodity. The application solves the problem that information cannot be extracted due to the fact that a user does not send a picture or fill in text description, improves recognition accuracy, makes up for the incompleteness of user-filled content, provides more accurate and standardized commodity key information for downstream recommendation and search algorithms, effectively helps commodities to be more efficiently exposed, and improves the speed of sales.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and in particular to a method and system for mining structured information of goods in a second-hand e-commerce scenario. Background Technology

[0002] With the development of internet applications and the logistics industry, more and more merchants and consumers prefer to complete transactions on e-commerce platforms. E-commerce platforms gather a large number of merchants and their products. To better serve merchants and consumers, e-commerce platforms are constantly striving to improve the quality of product display and provide fast services. When a merchant releases a new product, to ensure it quickly enters the platform's categories, it needs to be categorized across multiple dimensions based on its product description, uploaded photos, and other information to obtain structured information. This allows for matching based on consumer search terms, increasing the product's exposure. Currently, most e-commerce platforms require merchants to fill out forms with specific categorization information when releasing new products, such as the product's category (e.g., home goods, clothing), and each category further subdivided into subcategories. For example, the clothing category includes subcategories like brand, size, and style. Similarly, the computer accessories category includes a monitor subcategory, which includes subcategories like brand, interface type, and size. The platform sets the dimensions for the newly released product based on the form submitted by the merchant. However, for secondhand e-commerce platforms, the aforementioned settings cannot provide adequate structured information to represent the products. This is because a large proportion of sellers on these platforms are individuals, and the products they sell are primarily personal items. When filling out forms, it is sometimes difficult to accurately classify the items, or even incorrectly. Relying solely on user-filled forms to generate structured information across various dimensions is not only inaccurate but also lacks sufficient dimensions. Therefore, most secondhand e-commerce platforms currently use a combination of manual processing and forms, manually verifying the basic classification information in the forms and the product information filled in by the users to obtain structured information. However, with the increasing number of platform users (individuals or merchants posting products) and a large number of new products being posted daily, the aforementioned semi-automatic verification method can no longer meet business requirements. Summary of the Invention

[0003] To address the technical problems existing in the prior art, this invention proposes a method and system for mining structured information of goods in the context of second-hand e-commerce, which is used to provide structured information of newly released goods.

[0004] To address the aforementioned technical problems, according to one aspect of the present invention, a method for mining structured product information in a second-hand e-commerce scenario is provided, comprising the following steps: acquiring product information; in response to the inclusion of text content in the product information, extracting information from the text content to obtain multiple attributes and corresponding attribute information; in response to the inclusion of product images in the product information, obtaining the product's item terms using an image classification model; replacing non-standardized words in the attribute information with standardized words that are synonyms; and combining the current product's item terms and attribute information to form the product's structured information.

[0005] In some embodiments, the step of extracting information from text content includes: matching the text content with a preset product attribute table according to a rule matching algorithm to obtain first structured attribute information, wherein the first structured attribute information includes attributes and corresponding attribute information.

[0006] In some embodiments, the step of extracting information from text content further includes: using all or the remaining text content after rule matching of the text content as input to a sequence labeling model to obtain second structured attribute information, the second structured attribute information including attributes and corresponding attribute information; normalizing the same attribute information in the first structured attribute information and the second structured attribute information; and merging the normalized first structured attribute information and the second structured attribute information to obtain the structured attribute information of the product.

[0007] In some embodiments, the step of extracting information from text content further includes: in response to the extracted attribute having an associated attribute, adding associated attributes and corresponding associated attribute information based on the attribute and the corresponding attribute information.

[0008] In some embodiments, the rule matching algorithm is the Aho-Corasick automaton algorithm.

[0009] In some embodiments, when extracting item terms from text content, the extracted item terms are used as item terms in the structured information.

[0010] In some embodiments, the method further includes: querying whether the structured information of the product includes an item term; in response to the absence of an item term in the structured information of the product, querying a historical product record table based on the attribute information in the structured information, wherein in some embodiments, the product record table includes the item term of the product and the corresponding multiple attribute information; and determining the item term of the product in the product record table that has the highest matching degree with the current attribute information as the item term of the current product.

[0011] In some embodiments, the method further includes: counting the number of times a product term co-occurs with one or more attribute information; and in response to the number of times a product term co-occurs with an attribute information being greater than a threshold, storing the attribute and its corresponding attribute information in a product attribute table.

[0012] To address the aforementioned technical problems, according to another aspect of the present invention, a product structured information mining system for secondhand e-commerce scenarios is provided. This system includes a product information acquisition module, an information extraction module, an image classification module, an information standardization module, and an information determination module. The product information acquisition module is configured to acquire product information. The information extraction module, connected to the product information acquisition module, is configured to extract information from the text content of the product information to obtain multiple attributes and corresponding attribute information. The image classification module, connected to the product information acquisition module, is configured to use an image classification model to obtain the product's item terms based on the product image in the product information. The information standardization module, connected to the information extraction module, is configured to replace non-standardized words in the attribute information with standardized words that are synonyms. The information determination module, connected to the information standardization module and the image classification module, is configured to combine the current product's item terms and attribute information to form the product's structured information.

[0013] In some embodiments, the information extraction module includes: a rule matching unit configured to match text content with a preset product attribute table according to a rule matching algorithm to obtain first structured attribute information, the first structured attribute information including attributes and corresponding attribute information; a sequence labeling model unit configured to use all or the remaining text content after rule matching as input to a sequence labeling model to obtain second structured attribute information, the second structured attribute information including attributes and corresponding attribute information; and a normalization unit connected to the rule matching unit and the sequence labeling model unit, configured to normalize the same attribute information in the first structured attribute information and the second structured attribute information.

[0014] In some embodiments, the information extraction module further includes an association unit connected to the normalization unit, configured to add associated attributes and corresponding associated attribute information based on the attribute and the corresponding attribute information in response to the obtained attribute having associated attributes.

[0015] In some embodiments, the information determination module further includes: an item term query unit configured to query whether the currently obtained structured information includes an item term; an item term selection unit connected to the item term query unit, configured to use the item term as the item term in the structured information when the attribute information includes an item term; and an item term estimation unit connected to the item term query unit, configured to query a historical product record table based on the attribute information in the structured information when there is no item term in the currently obtained structured information; and determining the item term of the product in the product record table that has the highest matching degree with the current attribute information as the item term of the current product. In some embodiments, the product record table includes the item term of the product and multiple corresponding attribute information.

[0016] In some embodiments, the system further includes: a statistics module configured to count the number of times a product's item term co-occurs with one or more of its attribute information; and a recording module connected to the statistics module, configured to store the attribute and its corresponding attribute information in a product attribute table when the number of times a product's item term co-occurs with an attribute information is greater than a threshold.

[0017] In the context of secondhand e-commerce, most sellers are individual sellers who generally lack sufficient knowledge about the products they sell. This leads to brief descriptions when listing products, failing to accurately and comprehensively describe the goods, and often resulting in grammatical and vocabulary errors, as well as low-quality photos. This invention utilizes deep learning algorithms to extract key information from product information, compensating for these deficiencies. It provides downstream recommendation and search algorithms with more accurate and standardized key product information, effectively helping products gain more efficient exposure and increasing sales speed. This invention combines bimodal information from images and text during information extraction, solving the problem of information extraction failure due to users not providing images or text descriptions. The bimodal information also complements each other, improving recognition accuracy. Furthermore, this invention can utilize product attribute relationships extracted from past products to fill in missing product information, addressing the issue of sellers' insufficient knowledge and inaccurate descriptions of the products they sell in the secondhand market. Attached Figure Description

[0018] The preferred embodiments of the present invention will now be described in further detail with reference to the accompanying drawings, wherein:

[0019] Figure 1 This is a principle block diagram of a product structured information mining system in a second-hand e-commerce scenario provided by an embodiment of the present invention;

[0020] Figure 2 This is a flowchart of a method for mining structured product information in a second-hand e-commerce scenario according to an embodiment of the present invention;

[0021] Figure 3 This is a schematic diagram of product information according to an embodiment of the present invention;

[0022] Figure 4 This is a block diagram illustrating the principle of an information extraction module according to an embodiment of the present invention.

[0023] Figure 5 This is a simplified structural diagram of a BiLSTM-CNN-CRF neural network model according to an embodiment of the present invention;

[0024] Figure 6 This is a flowchart of information extraction from text content according to an embodiment of the present invention;

[0025] Figure 7 This is a schematic diagram of an information determination module according to an embodiment of the present invention;

[0026] Figure 8 This is a schematic diagram of a product structured information mining system in a second-hand e-commerce scenario provided by another embodiment of the present invention; and

[0027] Figure 9 This is a flowchart of a method for mining structured product information in a second-hand e-commerce scenario according to another embodiment of the present invention. Detailed Implementation

[0028] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, 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, 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.

[0029] In the following detailed description, reference can be made to the accompanying drawings, which form part of this application and illustrate specific embodiments of the present application. In the drawings, similar reference numerals describe substantially similar components in different figures. Specific embodiments of the present application are described in sufficient detail below to enable those skilled in the art to implement the technical solutions of the present application. It should be understood that other embodiments may also be utilized, or structural, logical, or electrical changes may be made to the embodiments of the present application.

[0030] Figure 1 This is a principle block diagram of a product structured information mining system in a second-hand e-commerce scenario according to an embodiment of the present invention, which includes a product information acquisition module 1, an information extraction module 2, an image classification module 3, an information standardization module 4, and an information determination module 5. Figure 2This is a flowchart illustrating a method for mining structured product information in a secondhand e-commerce scenario according to an embodiment of the present invention. Combined with... Figure 1 The product structured information mining method described in this embodiment is explained as follows:

[0031] Step S1a: Obtain information on newly published products. See also... Figure 1 The product information acquisition module 1 is used to acquire product information, which can be product information already stored in the platform's data without structured information mining, or newly released product information. In one embodiment, the second-hand e-commerce platform of this invention monitors user operations on the platform. When it detects that a user has released a new product, it sends a notification to this system. The product information acquisition module 1 reads the user's operation log according to the notification and retrieves the newly released product information, which includes text content and, in most cases, product images, such as... Figure 3 As shown, the obtained product information includes text content and images.

[0032] Step S2a involves extracting information from the text content to obtain multiple attributes and their corresponding attribute information. After reading the product information, the product information acquisition module 1 sends notifications to the information extraction module 2 and the image classification module 3, respectively, based on the type of content read. The information extraction module 2 extracts the corresponding attribute information from the text content according to a certain algorithm. In one embodiment, such as... Figure 4 As shown, the information extraction module 2 includes a rule matching unit 21, a sequence labeling model unit 22, and a normalization unit 23. The rule matching unit 21 matches the text content with a pre-set product attribute table according to a rule matching algorithm to obtain first structured attribute information. This first structured attribute information includes attributes and their corresponding attribute information. For example, the rule matching unit 21 uses an AC automaton to match each character of the text in the product information with the pre-set product attribute table. The pre-set product attribute table includes words and their corresponding attributes, as shown in Table 1 below. When a corresponding word is matched in the product attribute table, its attribute is obtained. Therefore, the rule matching unit 21 can match attributes to all or part of the text content, such as... Figure 3 The text content in the middle, after being matched by the rule matching unit 21, can be obtained as: {Condition: 95% new, Channel: Mainland China version, Standard: China Mobile 4G, China Unicom 4G, China Telecom 4G}. Among them, "Condition", "Channel", and "Standard" are attributes, while "95% new", "Mainland China version", and "China Mobile 4G, China Unicom 4G, China Telecom 4G" are the corresponding attribute information.

[0033] Table 1

[0034] calligraphy brush Item words Duan inkstone Item words cell phone Item words ……. …… Crane brand Oral-B brand ……. ……. Obsidian Grey color red color ……. …….

[0035] The sequence labeling model unit 22 takes all the text content in the product information or the remaining text content after matching as input to the sequence labeling model to obtain the second structured attribute information. The second structured attribute information includes attributes and corresponding attribute information. The sequence labeling model is, for example, a BiLSTM-CNN-CRF neural network model. Figure 5 The diagram shows a simplified structure of the BiLSTM-CNN-CRF neural network model. The text input layer feeds the model with the text content to be processed, such as "95% new iPhone X 64G Gray Chinese version". The next layer, for example, is a BERT model, used to segment the text into characters and obtain character vectors. These character vectors are then fed to the BiLSTM and CNN layers for processing. The BiLSTM layer combines two sets of LSTM layers with opposite learning directions (one in sentence order, one in reverse order) to perform sequence labeling on the input character vectors and output them to the CRF layer. Additionally, the CNN layer extracts local features of the current word and also outputs them to the CRF layer. The CRF layer merges the outputs of the BiLSTM and CNN layers, learning the optimal label sequence over the entire sequence, and finally outputting a corresponding type for each character, such as "9->b condition", "5->i_ condition", "new->i_ condition", "i->b_ model", "P->i_ model", "h->i_ model", "o->i_ model", "n->i_ model", "e->i_ model", etc. Finally, characters of the same type are merged to get "Condition: 95% new, Model: iPhone X, Capacity: 64G, Color: Gray, Channel: China version".

[0036] The normalization unit 23 is connected to the rule matching unit 21 and the sequence labeling model unit 22, and is used to normalize the same attribute information in the first and second structured attribute information. For example, the first structured information extracted from the text content by the rule matching unit 21 is: {condition: 95% new, channel: mainland China version, standard: China Mobile 4G, China Unicom 4G, China Telecom 4G}. The second structured information annotated from the text content by the sequence labeling model unit 22 is: {condition: 95% new, model: OnePlus 7, capacity: 256g, color: Obsidian Gray}. The normalization unit 23 checks the content of each attribute field one by one and compares it with the preset normalization table. When it finds the content in the normalization table, it replaces the content of the current attribute field with the corresponding content in the normalization table. For example, the content of the "system type" attribute field in the first structured information is "China Mobile 4G, China Unicom 4G, China Telecom 4G", which is replaced by the normalization unit 23 with "All Networks Compatible". The content of the "color" attribute field in the second structured information is "Obsidian Gray", which is replaced by the normalization unit 23 with "Gray".

[0037] In another embodiment, since certain attributes are related, when an attribute is obtained from text content but no related attribute is found, this embodiment can obtain the related attribute and attribute information based on the association relationship and attribute information of the specific attribute. Therefore, in this embodiment, the information extraction module 2 further includes an association unit 24, which queries the currently obtained attribute to determine whether it has a specific attribute. If so, it obtains the attribute associated with the specific attribute based on the attribute association relationship, referred to here as the associated attribute, and obtains the attribute information of the associated attribute based on the attribute information of the specific attribute. For example, in the aforementioned embodiment, the attribute "model" has a specific association relationship with "brand" when the product item term is "mobile phone". Therefore, when the attribute "model" is obtained from text content, "brand: OnePlus" can be obtained based on "model: OnePlus 7". This is then added to the structured information to supplement the missing content in the product information input by the user.

[0038] In order to comprehensively and quickly obtain all attributes and attribute information contained in the text content, in one embodiment, when extracting information from the text content, such as... Figure 6 As shown, it includes the following steps:

[0039] Step S21a: Obtain the first structured attribute information from the text content according to rule matching. Specifically, the rule matching unit 21 matches each character in the text content with the product attribute table from beginning to end according to a rule matching algorithm, such as an AC automaton. When a word is matched in the product attribute table, the attribute corresponding to the word is obtained, thereby obtaining the first structured attribute information.

[0040] Step S22a: Determine whether all the text content of the product has been matched. If it has been matched, proceed to step S24a; otherwise, proceed to step S23a.

[0041] In step S23a, the remaining text content is input into the model in the sequence labeling model unit 22, and the model performs sequence labeling to obtain the second structured attribute information.

[0042] Step S24a: Normalize and merge the same attribute information in the first structured attribute information and / or the second structured attribute information.

[0043] Step S25a: Query whether the attributes in the structured attribute information have associated attributes. If so, add the associated attributes and corresponding associated attribute information in step S26a; otherwise, end the process.

[0044] Of course, in step S23a, all the text content can also be input into the model in sequence labeling model unit 22, and then the same attributes and their field contents can be merged at the end.

[0045] Step S3a involves using an image classification model to obtain the product's item terms from the product images in the product information. Typically, when users on secondhand e-commerce platforms post items for sale, they include at least one product image in addition to a text description. Therefore, in this step, the image classification module 3 uses an image classification model in image processing to process the product image to obtain the product's item terms. The image classification model can be, for example, VGG, GoogleNet, or ResNet. The product's item terms are obtained after processing by the model. For example, the VGG model can be used to... Figure 3 The product image in the image is processed to obtain the item name "mobile phone".

[0046] Step S4a: Replace the corresponding attribute information with standardized information. Since users use diverse and highly personalized language when describing their products, this personalized vocabulary can hinder the system's search and matching process. Therefore, to reduce subsequent business complexity, after obtaining the user's structured product information, standardized vocabulary is used to replace synonymous personalized vocabulary. The system stores a thesaurus, which includes multiple synonyms expressing the same meaning, one of which is a standardized vocabulary representing those synonyms. For example, "Apple X" is a standardized vocabulary for "iphone X" and "phone X". The information standardization module 4 queries the thesaurus based on the vocabulary representing attribute information in the attribute fields. When a vocabulary is found in the thesaurus, it compares whether the vocabulary is a standardized vocabulary. If not, the standardized vocabulary replaces the vocabulary representing attribute information in the attribute fields. By traversing the vocabulary representing attribute information in the currently obtained attribute fields, non-standardized vocabulary is replaced with corresponding synonymous standardized vocabulary.

[0047] Step S5a: Combine the item terms and attribute information of the current product to form the structured information of the product. The information determination module 5 is connected to the information standardization module 4 and the image classification module 3, and combines the attribute information obtained after replacement by the information standardization module 4 with the item terms obtained by the image classification module 3 to form the structured information of the product. Since the attributes obtained by the information standardization module 4 may also be item terms, the information determination module 5 determines the item term to be used based on the current attributes and attribute information and the item terms obtained by the image classification module 3. Therefore, in one embodiment, as... Figure 7As shown, the information determination module 5 includes an item word query unit 51 and an item word selection unit 52. The item word query unit 51 queries whether the currently obtained information includes an item word, for example, it queries whether the attribute extracted from the text content includes the item word attribute. The item word selection unit 52 receives the query result. If the attribute extracted from the text content includes the item word attribute, the item word extracted from the text content is used as the item word in the final structured information. If the attribute is not extracted from the text content, the item word obtained from the image classification module 3 is used as the item word in the final structured information, thus finally obtaining the structured information of the product, which includes the item word and various attribute information. For Figure 3 The final structured information obtained is {item keywords: mobile phone, condition: 95% new, brand: OnePlus, model: OnePlus 7, capacity: 256g, memory: 8g, channel: Chinese version, network standard: full network compatibility, color: gray}.

[0048] In another embodiment, when the image classification module 3 fails to obtain item terms, or when there are no product images in the user's product information, and no item terms are extracted from the text, although structured attribute information has been extracted, item terms are missing. In this case, the present invention infers item terms from a historical product record table. The product record table, as shown in Table 2, includes product IDs, item terms, and corresponding attributes and attribute information from the second-hand platform.

[0049] Table 2

[0050] Product ID Item words Attributes and attribute information 1124961730565537805 cell phone Brand: Apple, Model: iPhone X, Capacity: 64GB, ... 1268115140793117543 Platform PC Brand: Apple, Model: iPad Pro 2019, Capacity: 256GB, ...

[0051] The information determination module 5 further includes an item word estimation unit 53, which is used to query the historical product record table based on the attribute information in the structured information when the currently obtained information does not include an item word, such as Table 2, compare the current attribute or attribute information with the attributes and corresponding attribute information in the product record table in Table 2, and determine the item word of the product with the highest attribute information matching degree as the item word of the current product.

[0052] Figure 8This is a block diagram illustrating the principle of a structured information mining system for goods in a second-hand e-commerce scenario according to another embodiment of the present invention. In this embodiment, in addition to the product information acquisition module 1, information extraction module 2, image classification module 3, information standardization module 4, and information determination module 5 included in the preceding embodiments, it also includes a statistics module 6 and a recording module 7. The statistics module 6 is used to count the co-occurrence frequency of product terms and one or more attribute information. For example, the product attributes "brand" and "model" appear in almost every product with the product term "mobile phone," thus these two attributes are two important descriptive dimensions for products with the product term "mobile phone." Similarly, for products with the product term "T-shirt," "size" and "material" are important descriptive dimensions. Therefore, these attributes and their corresponding attribute information are stored in the product attribute table shown in Table 1. The statistics module 6 counts the co-occurrence frequency of all product attributes and their attribute information and sends the statistical results to the recording module 7. When the co-occurrence frequency of a product's item term and an attribute and its information in the statistical results exceeds a threshold, the recording module 7 stores the item term's attribute and information in the product attribute table shown in Table 1. This expands the content of the product attribute table and provides a richer matching foundation for mining product structured information. Additionally, the recording module 7 also stores the currently obtained product structured information in the product and its attribute table shown in Table 2.

[0053] Figure 9 This is a flowchart of a method for mining structured product information in a second-hand e-commerce scenario according to another embodiment of the present invention. In this embodiment, the method includes the following steps:

[0054] Step S1b: Obtain information on newly released products.

[0055] Step S2b involves identifying the type of product information. If the product information includes text content, step S3b extracts information from the text content to obtain structured attribute information. If the product information includes images, step S4b uses image classification to process the product image and obtain item terms.

[0056] Step S5b: Combine the results obtained from steps S3b and S4b.

[0057] Step S6b: Determine if the current result includes item terms. If it does, in step S11b, determine if the number of item terms is one. If it is one, proceed to step 10b. If it is two, in step S12b, determine if the two item terms are the same, i.e., whether the item terms extracted from the text are consistent with the item terms obtained from the image. If they are consistent, in step S13b, delete the duplicate item terms. If they are inconsistent, it indicates that the text and image do not match, and in step S14b, send a notification to the user informing them that the current text description does not match the image. Step S15b monitors the user's actions on the product information, and in step S16b, determine if the user has modified the product information. If the user has modified the product information, proceed to step S1b and re-perform the structured information mining process. If not, proceed to step S15b to continue monitoring user behavior. If the current result does not contain item terms, proceed to step S7b.

[0058] Step S7b: Query the attribute information in the product record table.

[0059] Step S8b: Calculate the similarity between the current attributes and attribute information and the attributes and attribute information in the product record table.

[0060] Step S9b: Use the item word of the product with the highest similarity as the item word of the current product.

[0061] Step S10b: Record the structured information of the current product into the product record table according to the field format.

[0062] This embodiment combines bimodal information from images and text, which not only improves recognition accuracy but also enables timely detection of problems, allowing users to correct errors promptly.

[0063] The above embodiments are for illustrative purposes only and are not intended to limit the invention. Those skilled in the art can make various changes and modifications without departing from the scope of the invention. Therefore, all equivalent technical solutions should also fall within the scope of the invention.

Claims

1. A method for mining structured product information in a second-hand e-commerce scenario, including: Obtain information on second-hand goods; In response to the inclusion of text content in the information of second-hand goods, information extraction is performed on the text content to obtain multiple attributes and corresponding attribute information. The information extraction includes: matching the text content with a preset product attribute table according to a rule matching algorithm to obtain first structured attribute information; and using all or the remaining text content after rule matching as input to a sequence labeling model to obtain second structured attribute information; and normalizing and merging the same attribute information in the first structured attribute information and the second structured attribute information. In response to the inclusion of product images in the second-hand goods information, the item terms of the product are obtained using an image classification model; Replace non-standardized terms in the attribute information with their synonyms in the standardized terms; and Combine the item terms and attribute information of the current product to form the structured information of the product; Determine whether the structured information of the product includes item terms. If there is only one item term, record the structured information of the current product in the product record table according to the field format. If there are two item terms, determine whether the item terms extracted from the text are consistent with the item terms obtained from the image. If they match, the duplicate item terms are deleted; if they do not match, it means the text and image do not match, so a notification is sent to the user to inform them that the current text description does not match the image, and the user's actions on the second-hand goods information are monitored to determine whether the user has modified the second-hand goods information. If the user modifies the information of the second-hand goods, the process of mining the structured information of the goods will be repeated; if the information is not modified, the user behavior will continue to be monitored. If the current result does not contain an item term, in response to the absence of an item term in the product's structured information, a historical product record table is queried based on the attribute information in the product's structured information. This product record table includes the product's item term and corresponding attribute information. The item word of the product with the highest matching degree with the current attribute information in the product record table is determined as the item word of the current product.

2. The method according to claim 1, wherein the step of extracting information from the text content further includes: In response to the fact that the extracted attribute has associated attributes, associated attributes and corresponding associated attribute information are added based on the attribute and the corresponding attribute information.

3. The method according to claim 1, wherein the rule matching algorithm is the AC automaton algorithm.

4. According to the method of claim 1, when extracting item words from the text content, the extracted item words are used as item words in the structured information of the product.

5. The method of claim 1, further comprising: Count the number of times the item term and one or more attribute information of the current product co-occur. as well as If the number of times an item term and an attribute information co-occurs for the current product is greater than a threshold, the attribute and its corresponding attribute information are stored in the product attribute table.

Images