Commodity matching method and device and storage medium

Abstract

The embodiment of the invention discloses a commodity matching method and device and a storage medium, and the method comprises the steps: obtaining a to-be-searched text, carrying out the text recall processing of the to-be-searched text, obtaining a plurality of commodity recall texts, and enabling each commodity recall text to correspond to a commodity text type; according to the commodity text type corresponding to each commodity recall text, adding a commodity type label for each commodity recall text to obtain a labeled commodity recall text; performing text splicing on all the labeled commodity recall texts to obtain a to-be-processed commodity text, and performing feature extraction on the to-be-processed commodity text to obtain commodity text feature information of each labeled commodity recall text; obtaining candidate commodity information preset in a candidate commodity link library according to the to-be-searched text, and obtaining multiple pieces of commodity feature information of the candidate commodity information; and performing text matching processing on all the commodity text feature information and all the commodity feature information to obtain a matching result of the to-be-searched text and the candidate commodity information.

Description

Technical Field

[0001] The present invention relates to the field of computer technology, and in particular to a commodity matching method, device and storage medium. Background Art

[0002] At present, e-commerce has become one of the most important application services on the Internet. There is a large amount of product information in the text information on Internet web pages, such as search web pages, news, blogs, etc., but most web pages themselves do not have relevant descriptions and introductions about these products. In order to help people better understand the content of product information, many websites or authors will link the products appearing on the web pages to the corresponding product entries in the product library of the e-commerce platform to provide readers with more detailed product information. This process of retrieving and matching similar product entries from the product library based on text information and establishing a link relationship between the text information and the product entry is called chain pointing. Since the text information to be matched includes different types of text, and the meanings of different types of text are different and highly different, the method used in the related art to directly retrieve and match the text information to be matched from the product library will ignore the differences between different types of text, affect the accuracy of retrieval and matching, and result in poor chain pointing effect. Summary of the invention

[0003] The following is a summary of the subject matter of the detailed description of the present invention. This summary is not intended to limit the scope of the claims.

[0004] The embodiments of the present invention provide a commodity matching method, device and storage medium, which can improve the accuracy of commodity matching and improve the chain finger effect.

[0005] In one aspect, an embodiment of the present invention provides a product matching method, comprising the following steps:

[0006] Acquire a text to be searched, perform text recall processing on the text to be searched, and obtain a plurality of product recall texts, each of which corresponds to a product text type;

[0007] According to the commodity text type corresponding to each of the commodity recall texts, a commodity type label is added to each of the commodity recall texts to obtain a labeled commodity recall text;

[0008] Performing text splicing on all the marked product recall texts to obtain the product texts to be processed, and performing feature extraction on the product texts to be processed to obtain product text feature information of each of the marked product recall texts;

[0009] Acquire candidate product information preset in a candidate product link library according to the text to be searched, and acquire multiple product feature information of the candidate product information;

[0010] Text matching processing is performed on all the commodity text feature information and all the commodity feature information to obtain a matching result between the text to be searched and the candidate commodity information.

[0011] On the other hand, an embodiment of the present invention further provides a commodity matching device, characterized by comprising:

[0012] A text recall unit, used to obtain a text to be searched, perform text recall processing on the text to be searched, and obtain a plurality of product recall texts, each of which corresponds to a product text type;

[0013] a label adding unit, configured to add a product type label to each of the product recall texts according to the product text type corresponding to each of the product recall texts, so as to obtain a marked product recall text;

[0014] A feature extraction unit is used to perform text splicing on all the marked product recall texts to obtain a product text to be processed, and perform feature extraction on the product text to be processed to obtain product text feature information of each of the marked product recall texts;

[0015] An information acquisition unit, configured to acquire candidate commodity information preset in a candidate commodity link library according to the text to be searched, and acquire a plurality of commodity feature information of the candidate commodity information;

[0016] The text matching unit is used to perform text matching processing on all the commodity text feature information and all the commodity feature information to obtain a matching result between the text to be searched and the candidate commodity information.

[0017] Optionally, the text recall unit is further used for:

[0018] Performing semantic feature recognition on the text to be searched to obtain multiple semantic feature information;

[0019] Text recall processing is performed according to each of the semantic feature information to obtain multiple product recall texts.

[0020] Optionally, the text recall unit is further used for:

[0021] Performing text recall processing according to each of the semantic feature information to obtain multiple candidate product recall texts;

[0022] Obtaining multiple supplementary product recall texts according to the candidate product recall text and the candidate product information;

[0023] A plurality of product recall texts are obtained according to all the candidate product recall texts and all the supplementary product recall texts.

[0024] Optionally, the text recall unit is further used for:

[0025] Determining the type of supplementary product text corresponding to the candidate product recall text according to the candidate product recall text and the candidate product information;

[0026] A plurality of supplementary product recall texts are obtained according to the supplementary product text type.

[0027] Optionally, the information acquisition unit is further used to:

[0028] According to the product type information, a candidate product type label is added to each candidate product sub-information to obtain the labeled product information;

[0029] All the marked commodity information is concatenated to obtain the commodity information to be processed, and the feature extraction is performed on the commodity information to be processed to obtain the commodity feature information of each of the marked commodity information;

[0030] The candidate product information includes a plurality of candidate product sub-information, and the candidate product sub-information corresponds to the product type information.

[0031] Optionally, the text matching unit is further used for:

[0032] Performing text matching processing on each of the commodity text feature information and all of the commodity feature information to obtain a feature matching value corresponding to each of the commodity text feature information;

[0033] According to all the feature matching values, a matching result between the text to be searched and the candidate product information is obtained.

[0034] Optionally, the text matching unit is further used for:

[0035] Performing similarity matching processing on each of the commodity text feature information and each of the commodity feature information to obtain feature similarities between each of the commodity text feature information and each of the commodity feature information;

[0036] For each of the commodity text feature information, the feature similarity with the largest corresponding value is used as the feature matching value corresponding to each of the commodity text feature information.

[0037] Optionally, the text matching unit is further used for:

[0038] Adding all the feature matching values ​​to obtain a matching result between the text to be searched and the candidate product information;

[0039] Alternatively, weighted addition processing is performed on all the feature matching values ​​to obtain a matching result between the text to be searched and the candidate product information.

[0040] Optionally, there are multiple candidate product information; the text matching unit is further used to:

[0041] Performing text matching processing on all the product text feature information and the product feature information corresponding to each of the candidate product information to obtain candidate matching results between the text to be searched and each of the candidate product information;

[0042] According to the size of each candidate matching result, each candidate product information is sorted to obtain the matching result between the text to be searched and the candidate product information.

[0043] Optionally, the text matching unit is further used for:

[0044] When the matching result is that the text to be searched does not match the candidate product information, and the product feature information corresponding to the product key information is consistent with the product text feature information corresponding to the text key type, adding an error correction mark to the text to be searched and the candidate product information;

[0045] According to the error correction mark, the text to be searched and the candidate product information are subjected to secondary matching processing to obtain a secondary matching result between the text to be searched and the candidate product information;

[0046] The commodity feature information corresponds to the commodity type information, the commodity type information includes the commodity key information, and the commodity text type includes the text key type.

[0047] Optionally, the text matching unit is further used for:

[0048] When the matching result is that the text to be searched does not match the candidate product information, and the product feature information corresponding to the product key information is consistent with the product text feature information corresponding to the text key type, adding an error correction mark to the text to be searched and the candidate product information;

[0049] According to the error correction mark, the text to be searched and the candidate product information are subjected to secondary matching processing to obtain a secondary matching result between the text to be searched and the candidate product information;

[0050] The commodity feature information corresponds to the commodity type information, the commodity type information includes the commodity key information, and the commodity text type includes the text key type.

[0051] On the other hand, an embodiment of the present invention further provides a commodity matching device, including:

[0052] at least one processor;

[0053] at least one memory for storing at least one program;

[0054] When at least one of the programs is executed by at least one of the processors, the commodity matching method as described above is implemented.

[0055] On the other hand, an embodiment of the present invention further provides a computer-readable storage medium, which stores a computer program executable by a processor. When the computer program executable by the processor is executed by the processor, it is used to implement the commodity matching method as described above.

[0056] On the other hand, an embodiment of the present invention further provides a computer program product, including a computer program or computer instructions, wherein the computer program or the computer instructions are stored in a computer-readable storage medium, and a processor of a product matching device reads the computer program or the computer instructions from the computer-readable storage medium, and the processor executes the computer program or the computer instructions, so that the product matching device performs the product matching method as described above.

[0057] The embodiment of the present invention includes at least the following beneficial effects: after obtaining the text to be searched, the text to be searched is processed for text recall to obtain multiple commodity recall texts, each of which has a corresponding commodity text type, so that text information of different commodity text types can be obtained, thereby improving the richness of the information to be matched. By using the commodity text types of the commodity recall texts, commodity type labels are added to each commodity recall text to obtain the marked commodity recall text, thereby distinguishing the text information of different commodity text types. Since the text content of different commodity text types has different importance for commodity information, by distinguishing the information of different importance of the text to be searched, the difference of each feature in the text to be searched is highlighted, which helps to improve the matching accuracy of the text to be searched and the candidate commodity information. The commodity text to be processed is obtained by text splicing all the marked commodity recall texts, and feature extraction is performed to obtain the commodity text feature information of each marked commodity recall text, so that the obtained commodity text feature information can retain the difference between different marked commodity recall texts, thereby improving the matching accuracy. By obtaining preset candidate product information and corresponding multiple product feature information based on the text to be searched, relevant candidate product information can be preliminarily screened for product matching, and at the same time, the amount of data processing for the candidate product information can be reduced, which helps to improve the accuracy and efficiency of product matching. By performing text matching processing on the product text feature information and the product feature information, the matching result between the text to be searched and the candidate product information can be determined, and then the product information associated with the text to be searched can be determined. Therefore, by adding product type tags to the product recall text recalled by the text to be searched before product matching according to the respective product text types, it is possible to distinguish information of different importance in the text to be searched, improve the difference in features during product matching, thereby improving the accuracy of product matching and improving the product chain pointing effect.

[0058] Other features and advantages of the present invention will be described in the following description, and partly become apparent from the description, or understood by practicing the present invention. The purpose and other advantages of the present invention can be realized and obtained by the structures particularly pointed out in the description, claims and drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] The accompanying drawings are used to provide a further understanding of the technical solution of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the technical solution of the present invention and do not constitute a limitation on the technical solution of the present invention.

[0060] Figure 1 is a schematic diagram of an implementation environment provided by an embodiment of the present invention;

[0061] Figure 2 is a schematic diagram of an application scenario provided by an embodiment of the present invention;

[0062] Figure 3 is a schematic diagram of another implementation environment provided by an embodiment of the present invention;

[0063] Figure 4 is a schematic diagram of another application scenario provided by an embodiment of the present invention;

[0064] Figure 5 is a flow chart of a commodity matching method provided by an embodiment of the present invention;

[0065] Figure 6 is a schematic diagram of a text recall process provided by an embodiment of the present invention;

[0066] Figure 7 is a schematic diagram of a text recall process provided by another embodiment of the present invention;

[0067] Figure 8 is a schematic diagram of a text recall process provided by another embodiment of the present invention;

[0068] Fig. 9 It is a schematic diagram of a process of splicing all the recalled texts of marked commodities to form the commodity text to be processed provided by an embodiment of the present invention;

[0069] Fig.10 is a schematic diagram of a process for obtaining commodity feature information provided by an embodiment of the present invention;

[0070] Fig.11 is a schematic diagram of a process for obtaining a matching result between a text to be searched and candidate product information provided by an embodiment of the present invention;

[0071] Fig.12 It is a schematic diagram of a process of obtaining a matching result between a text to be searched and multiple candidate product information provided by an embodiment of the present invention;

[0072] Fig.13 is a process schematic diagram of a commodity matching method provided by an embodiment of the present invention;

[0073] Fig.14 is a flowchart of steps of a commodity matching method provided by a specific example of the present invention;

[0074] Fig.15 is a schematic diagram of a commodity matching device provided by an embodiment of the present invention;

[0075] Fig.16 It is a schematic diagram of a commodity matching device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0076] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0077] In the following description, reference is made to “some embodiments”, which describe a subset of all possible embodiments, but it will be understood that “some embodiments” may be the same subset or different subsets of all possible embodiments and may be combined with each other without conflict.

[0078] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by those skilled in the art to which the present invention belongs. The terms used herein are only for the purpose of describing the embodiments of the present invention and are not intended to limit the present invention.

[0079] To facilitate understanding of the technical solution provided by the embodiment of the present invention, some nouns and terms used in the embodiment of the present invention are explained here:

[0080] 1) Artificial Intelligence (AI) is the theory, method, technology and application system that uses digital computers or machines controlled by digital computers to simulate, extend and expand human intelligence, perceive the environment, acquire knowledge and use knowledge to obtain the best results. In other words, artificial intelligence is a comprehensive technology in computer science that attempts to understand the essence of intelligence and produce a new intelligent machine that can respond in a similar way to human intelligence. Artificial intelligence is to study the design principles and implementation methods of various intelligent machines so that machines have the functions of perception, reasoning and decision-making. Artificial intelligence technology is a comprehensive discipline that covers a wide range of fields, including both hardware-level technology and software-level technology. Basic artificial intelligence technologies generally include technologies such as sensors, dedicated artificial intelligence chips, cloud computing, distributed storage, big data processing technology, operation / interaction systems, and mechatronics. Artificial intelligence software technology mainly includes computer vision technology, speech processing technology, natural language processing technology, and machine learning / deep learning.

[0081] 2) Natural language processing (NLP) is an important direction in the fields of computer science and artificial intelligence. It studies various theories and methods that can achieve effective communication between people and computers using natural language. Natural language processing is a science that integrates linguistics, computer science, and mathematics. Therefore, research in this field will involve natural language, that is, the language people use in daily life, so it is closely related to the study of linguistics. Natural language processing technology usually includes text processing, semantic understanding, machine translation, robot question answering, knowledge graph and other technologies.

[0082] 3) Machine Learning (ML) is a multi-disciplinary subject involving probability theory, statistics, approximation theory, convex analysis, algorithm complexity theory and other disciplines. It specializes in studying how computers simulate or implement human learning behavior to acquire new knowledge or skills and reorganize existing knowledge structures to continuously improve their performance. Machine learning is the core of artificial intelligence and the fundamental way to make computers intelligent. Its applications are spread across all areas of artificial intelligence. Machine learning and deep learning usually include artificial neural networks, belief networks, reinforcement learning, transfer learning, inductive learning, and self-learning.

[0083] 4) Standard Product Unit (SPU) is the smallest unit of commodity information aggregation, a set of reusable and easily searchable standardized information that describes the characteristics of a product. In layman's terms, commodities with the same attribute values ​​and characteristics can be called an SPU.

[0084] 5) Chain finger, which can refer to retrieving similar SPU products from the complete SPU library based on product information or search terms or key information extracted from reading articles, and binding the corresponding relationship between the two.

[0085] At present, e-commerce has become one of the most important application services on the Internet. There is a large amount of product information in the text information on Internet web pages, such as search web pages, news, blogs, etc., but most web pages themselves do not have relevant descriptions and introductions about these products. In order to help people better understand the content of product information, many websites or authors will link the products appearing on the web pages to the corresponding product entries in the product library of the e-commerce platform to provide readers with more detailed product information. This process of retrieving and matching similar product entries from the product library based on text information and establishing a link relationship between the text information and the product entry is called chain pointing. Since the text information to be matched includes different types of text, and the meanings of different types of text are different and highly different, and the importance of the meanings of different types of texts is also different, such as the importance of brands and titles. In the related art, the method of directly retrieving and matching the text information to be matched from the product library will ignore the differences between different types of texts and the importance of the meanings of different types of texts, which will affect the accuracy of retrieval and matching, resulting in poor chain pointing effect. In addition, since the text information to be matched is in various forms, and the product entries in the product library use a standard fixed form to store product text information, if the form of the text information to be matched is inconsistent with the form of the product entry, it will also affect the accuracy of product matching.

[0086] In order to solve the problem of poor product matching between the text information to be matched and the product terms in the product library, improve the accuracy of product matching, and improve the linking effect, the embodiment of the present invention provides a product matching method, a product matching device, a computer-readable storage medium, and a computer program product. After obtaining the text to be searched, the text to be searched is processed for text recall to obtain multiple product recall texts, each of which has a corresponding product text type, so that text information of different product text types can be obtained, and the richness of the information to be matched can be improved. Then, the product text type of each product recall text can be used to add a product type label to each product recall text to obtain a labeled product recall text, so that the text information of different product text types can be distinguished. Since the text content of different product text types has different importance for product information, by distinguishing the information of different importance of the text to be searched, the differences of each feature in the text to be searched are highlighted, which helps to improve the matching accuracy of the text to be searched and the candidate product information. Then, the product text to be processed can be obtained by text splicing of all the marked product recall texts, and feature extraction can be performed to obtain the product text feature information of each marked product recall text, so that the obtained product text feature information can retain the differences between different marked product recall texts and improve the matching accuracy. According to the text to be searched, the preset candidate product information and the corresponding multiple product feature information are obtained, and the relevant candidate product information can be preliminarily screened for product matching, and the data processing amount for the candidate product information can be reduced, which is helpful to improve the accuracy and matching efficiency of product matching. By performing text matching processing on the product text feature information and the product feature information, the matching result between the text to be searched and the candidate product information can be determined, and then the product information associated with the text to be searched can be determined. Therefore, by adding product type tags to the product recall text recalled by the text to be searched before product matching according to the respective product text types, it is possible to distinguish information of different importance in the text to be searched, improve the difference of features during product matching, thereby improving the accuracy of product matching and improving the product chain pointing effect.

[0087] The method provided by the embodiment of the present invention can be applied to different technical fields, including but not limited to artificial intelligence, natural language processing and other technical fields.

[0088] Reference Figure 1 , Figure 1 1 is a schematic diagram of an implementation environment provided by an embodiment of the present invention. The implementation environment may include a first user terminal 101 and a first application server 102, and the first user terminal 101 and the first application server 102 may be directly or indirectly linked via wired or wireless communication. The first user terminal 101 and the first application server 102 may be nodes in a blockchain, which is not specifically limited in this embodiment.

[0089] The first application server 102 may be an independent physical server, or a server cluster or distributed system composed of multiple physical servers, or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, CDN (Content Delivery Network), and big data and artificial intelligence platforms. Among them, the first application server 102 may be deployed with a machine learning model based on artificial intelligence, and the machine learning model may be used to perform semantic recognition or retrieval matching on the data.

[0090] The first user terminal 101 may include but is not limited to a smart phone, a tablet, a computer, an intelligent voice interaction device, a smart home appliance, a vehicle-mounted terminal, an aircraft, etc.

[0091] Among them, the first application server 102 can be deployed with a product matching module, and a machine learning model based on artificial intelligence can be deployed in the product matching module. The product matching module has at least the function of using the machine learning model to perform semantic recognition on the data, and the function of matching and processing the feature data. For example, after receiving the text to be searched, the text to be searched can be firstly processed for text recall to obtain multiple product recall texts, wherein each product recall text corresponds to a product text type; then, according to the product text type corresponding to each product recall text, a product type label can be added to each product recall text to obtain a marked product recall text; then, all the marked product recall texts can be text spliced ​​to obtain the product text to be processed, and feature extraction can be performed on the product text to be processed to obtain product text feature information of each marked product recall text; in addition, the candidate product information preset in the candidate product link library can be obtained according to the text to be searched, and multiple product feature information of the candidate product information can be obtained; after obtaining the product feature information and the product text feature information, all the product text feature information and all the product feature information can be text matched to obtain the matching result of the text to be searched and the candidate product information. Therefore, by adding product type labels to the product recall texts recalled by the search texts before product matching according to the respective product text types, it is possible to distinguish information of different importance in the search texts and improve the difference in features during product matching, thereby improving the accuracy of product matching and improving the product chain pointing effect.

[0092] Reference Figure 2As shown, in an application scenario, it is assumed that the first user terminal 101 is a computer, and the first application server 102 is a cloud server deployed with a product matching module. In the process of product development, the user can input the text to be searched through the first user terminal 101, or the user can download and obtain network texts such as online articles and online news from the open source information source through the first user terminal 101, and extract the content of these network texts to obtain the text to be searched, or the text information (such as product information) pre-stored in the first user terminal 101 can be used as the text to be searched. Therefore, after the first user terminal 101 obtains the text to be searched, the first user terminal 101 can transmit the text to be searched to the first application server 102, so that the first application server 102 matches the search text with the product to achieve the purpose of linking. After the first application server 102 obtains the text to be searched, the first application server 102 can perform text recall processing on the search text to obtain multiple product recall texts. Since each product recall text corresponds to a product text type, text information of different product text types can be obtained, thereby improving the richness of the information to be matched. After obtaining the product recall text, the first application server 102 can add a product type tag to each product recall text according to the product text type corresponding to each product recall text to obtain a labeled product recall text; wherein, in the process of performing text recall processing on the search text to obtain multiple product recall texts, the first application server 102 can perform semantic feature recognition on the search text to obtain multiple semantic feature information, and then perform text recall processing according to each semantic feature information to obtain multiple product recall texts; and in the process of performing text recall processing according to each semantic feature information to obtain multiple product recall texts, the first application server 102 can perform text recall processing according to each semantic feature information to obtain multiple candidate product recall texts, and then obtain multiple supplementary product recall texts according to the candidate product recall texts and the candidate product information. Then, the first application server 102 can obtain multiple product recall texts according to all candidate product recall texts and all supplementary product recall texts; in addition, in the process of obtaining multiple supplementary product recall texts according to the candidate product recall text and the candidate product information, the first application server 102 can determine the supplementary product text type corresponding to the candidate product recall text according to the candidate product recall text and the candidate product information, and then can obtain multiple supplementary product recall texts according to the supplementary product text type; therefore, text information of different product text types can be distinguished. Since the importance of text content of different product text types to product information is different, by distinguishing information of different importance in the text to be searched, the differences of each feature in the text to be searched are highlighted, which helps to improve the matching accuracy between the text to be searched and the candidate product information.After obtaining the marked product recall text, the first application server 102 can concatenate all the marked product recall texts to obtain the product text to be processed, and then perform feature extraction on the product text to be processed to obtain product text feature information of each marked product recall text, so that the obtained product text feature information can retain the differences between different marked product recall texts and improve the matching accuracy.

[0093] After the first application server 102 obtains the text to be searched, the first application server 102 can obtain the candidate product information preset in the candidate product link library according to the text to be searched, and obtain multiple product feature information of the candidate product information; wherein the candidate product information includes multiple candidate product sub-information, and the candidate product sub-information corresponds to product type information. In the process of obtaining the product feature information, the first application server 102 can add a candidate product type label to each candidate product sub-information according to the product type information to obtain the marked product information, and then all the marked product information can be spliced ​​to obtain the product information to be processed, and the features of the product information to be processed can be extracted to obtain the product feature information of each marked product information, so that the relevant candidate product information can be preliminarily screened for product matching, and at the same time, the data processing amount for the candidate product information can be reduced, which is helpful to improve the accuracy and matching efficiency of product matching.

[0094] After the first application server 102 obtains the product text feature information and the product feature information, the first application server 102 may perform text matching processing on all product text feature information and all product feature information to obtain a matching result between the text to be searched and the candidate product information; wherein, in the process of performing text matching processing on all product text feature information and all product feature information to obtain a matching result between the text to be searched and the candidate product information, the first application server 102 may perform text matching processing on each product text feature information and all product feature information to obtain a feature matching value corresponding to each product text feature information, and then obtain a matching result between the text to be searched and the candidate product information according to all the feature matching values; and in the process of performing text matching processing on each product text feature information and all product feature information to obtain a matching result between the text to be searched and the candidate product information. In the process of obtaining the feature matching value corresponding to each commodity text feature information, the first application server 102 can perform similarity matching processing on each commodity text feature information and each commodity feature information to obtain the feature similarity between each commodity text feature information and each commodity feature information, and then for each commodity text feature information, the feature similarity with the largest corresponding value is used as the feature matching value corresponding to each commodity text feature information; in addition, in the process of obtaining the matching result of the to-be-searched text and the candidate commodity information according to all feature matching values, the first application server 102 can perform addition processing on all feature matching values ​​to obtain the matching result of the to-be-searched text and the candidate commodity information, or can perform weighted addition processing on all feature matching values ​​to obtain the matching result of the to-be-searched text and the candidate commodity information. Therefore, by adding commodity type tags to the commodity recall text recalled by the to-be-searched text according to the respective commodity text types before commodity matching, it is possible to distinguish information of different importance in the to-be-searched text, improve the difference of features during commodity matching, thereby improving the accuracy of commodity matching and improving the commodity chain pointing effect.

[0095] Since the commodity feature information corresponds to commodity type information, the commodity type information includes commodity key information, and the commodity text type includes text key type, and different types of information have different importance to the overall text information (i.e., the text to be searched and the candidate commodity information), after the first application server 102 obtains the matching result of the text to be searched and the candidate commodity information, the first application server 102 can match the commodity feature information corresponding to the commodity key information with the commodity text feature information corresponding to the text key type; when the matching result is that the text to be searched and the candidate commodity information do not match, and the commodity feature information corresponding to the commodity key information is consistent with the commodity text feature information corresponding to the text key type, an error correction mark is added to the text to be searched and the candidate commodity information, and then, the error correction mark can be used to match the text to be searched and the candidate commodity information. Error mark, perform secondary matching processing on the text to be searched and the candidate product information to obtain the secondary matching result of the text to be searched and the candidate product information; in addition, when the matching result is that the text to be searched and the candidate product information do not match, and the product feature information corresponding to the product key information is consistent with the product text feature information corresponding to the text key type, add an error correction mark to the text to be searched and the candidate product information, and then, according to the error correction mark, perform secondary matching processing on the text to be searched and the candidate product information to obtain the secondary matching result of the text to be searched and the candidate product information. By verifying the matching result, the text to be searched corresponding to the matching result with potential errors and the candidate product information are matched secondary, which helps to improve the accuracy of product matching and improve the product chain pointing effect.

[0096] Reference Figure 3 , Figure 3 A schematic diagram of another implementation environment provided by an embodiment of the present invention. The implementation environment may include a second user terminal 301, a second application server 302, and a third application server 303, and the second user terminal 301, the second application server 302, and the third application server 303 may be directly or indirectly linked via wired or wireless communication. Among them, the second user terminal 301, the second application server 302, and the third application server 303 may be nodes in a blockchain, and this embodiment does not specifically limit this.

[0097] The second user terminal 301 may include but is not limited to a smart phone, a tablet, a computer, an intelligent voice interaction device, a smart home appliance, a vehicle-mounted terminal, an aircraft, etc.

[0098] The second application server 302 and the third application server 303 can be independent physical servers, or server clusters or distributed systems composed of multiple physical servers, or cloud servers that provide basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, CDN (Content Delivery Network), and big data and artificial intelligence platforms. Among them, the second application server 302 has at least the functions of maintaining the normal operation of the system and storing the data information to be matched required in the commodity matching process, and the third application server 303 has at least the functions of processing the stored data information to be matched required in the commodity matching process. For example, the second application server 302 can be deployed with a candidate product link library, which includes multiple candidate product information, that is, the second application server 302 can save the candidate product information required for the product matching process, and the third application server 303 can, after receiving the text to be searched, first perform text recall processing on the text to be searched to obtain multiple product recall texts, wherein each product recall text corresponds to a product text type; then, according to the product text type corresponding to each product recall text, a product type label can be added to each product recall text to obtain a marked product recall text; then, all the marked product recall texts can be text spliced ​​to obtain the product text to be processed, and feature extraction can be performed on the product text to be processed to obtain product text feature information of each marked product recall text; in addition, the third application server 303 can call the candidate product information in the second application server 302 according to the text to be searched, and obtain multiple product feature information of the candidate product information; after obtaining the product feature information and the product text feature information, all product text feature information and all product feature information can be text matched to obtain the matching result of the text to be searched and the candidate product information. Therefore, by adding product type labels to the product recall texts recalled by the search texts before product matching according to the respective product text types, it is possible to distinguish information of different importance in the search texts and improve the difference in features during product matching, thereby improving the accuracy of product matching and improving the product chain pointing effect.

[0099] Reference Figure 4As shown, in another application scenario, it is assumed that the second user terminal 301 is a computer, the second application server 302 is a server running a product matching system, and the second application server 302 stores data information to be matched for product matching, which may include candidate product information in a candidate product link library, or text to be searched transmitted by the second user terminal 301; and the third application server 303 is a server for processing the data information to be matched. During the product operation process, the user can input the text to be searched through the second user terminal 301, or the user can download network texts such as online articles and online news from an open source information source through the second user terminal 301, and extract the content of these network texts to obtain the text to be searched, or the text information (such as product information) pre-stored in the second user terminal 301 can be used as the text to be searched. Therefore, after the second user terminal 301 obtains the text to be searched, the second user terminal 301 can transmit the text to be searched to the second application server 302, so that the second application server 302 can store the text to be searched, which is convenient for the third application server 303 to call the text to be searched for commodity matching, thereby achieving the purpose of linking. Therefore, the third application server 303 can call the text to be searched stored in the second application server 302, and perform text recall processing on the text to be searched to obtain multiple commodity recall texts. Since each commodity recall text corresponds to a commodity text type, text information of different commodity text types can be obtained, thereby improving the richness of the information to be matched.After obtaining the product recall text, the third application server 303 can add a product type tag to each product recall text according to the product text type corresponding to each product recall text, so as to obtain a labeled product recall text; wherein, in the process of performing text recall processing on the search text to obtain multiple product recall texts, the third application server 303 can perform semantic feature recognition on the search text to obtain multiple semantic feature information, and then perform text recall processing according to each semantic feature information to obtain multiple product recall texts; and in the process of performing text recall processing according to each semantic feature information to obtain multiple product recall texts, the third application server 303 can perform text recall processing according to each semantic feature information to obtain multiple candidate product recall texts, and then obtain multiple supplementary product recall texts according to the candidate product recall texts and the candidate product information. Supplementary product recall text, then, the third application server 303 can obtain multiple product recall texts according to all candidate product recall texts and all supplementary product recall texts; in addition, in the process of obtaining multiple supplementary product recall texts according to the candidate product recall text and the candidate product information, the third application server 303 can determine the supplementary product text type corresponding to the candidate product recall text according to the candidate product recall text and the candidate product information, and then obtain multiple supplementary product recall texts according to the supplementary product text type; therefore, the text information of different product text types can be distinguished. Since the text content of different product text types has different importance for product information, by distinguishing the information of different importance of the text to be searched, the difference of each feature in the text to be searched is highlighted, which helps to improve the matching accuracy of the text to be searched and the candidate product information. After obtaining the marked product recall text, the third application server 303 can perform text splicing of all marked product recall texts to obtain the product text to be processed, and then perform feature extraction on the product text to be processed to obtain the product text feature information of each marked product recall text, so that the obtained product text feature information can retain the difference between different marked product recall texts and improve the matching accuracy.

[0100] After the third application server 303 obtains the text to be searched in the second application server 302 through calling, the third application server 303 can call the candidate product information pre-stored in the candidate product link library of the second application server 302 according to the text to be searched, and call to obtain multiple product feature information of the candidate product information; wherein the candidate product information includes multiple candidate product sub-information, and the candidate product sub-information corresponds to product type information. In the process of obtaining the product feature information, the third application server 303 can first add a candidate product type label to each candidate product sub-information according to the product type information to obtain the marked product information, and then all the marked product information can be spliced ​​to obtain the product information to be processed, and the features of the product information to be processed are extracted to obtain the product feature information of each marked product information, so that the relevant candidate product information can be preliminarily screened for product matching, and at the same time, the data processing amount for the candidate product information can be reduced, which is helpful to improve the accuracy and matching efficiency of product matching.

[0101] After the third application server 303 obtains the product text feature information and the product feature information, the third application server 303 may perform text matching processing on all product text feature information and all product feature information to obtain a matching result between the text to be searched and the candidate product information; wherein, in the process of performing text matching processing on all product text feature information and all product feature information to obtain a matching result between the text to be searched and the candidate product information, the third application server 303 may perform text matching processing on each product text feature information and all product feature information to obtain a feature matching value corresponding to each product text feature information, and then, based on all feature matching values, obtain a matching result between the text to be searched and the candidate product information; and in the process of performing text matching processing on each product text feature information and all product feature information to obtain a matching result between the text to be searched and the candidate product information. In the process of obtaining the feature matching value corresponding to each commodity text feature information, the third application server 303 can perform similarity matching processing on each commodity text feature information and each commodity feature information to obtain the feature similarity between each commodity text feature information and each commodity feature information, and then for each commodity text feature information, the feature similarity with the largest corresponding value is used as the feature matching value corresponding to each commodity text feature information; in addition, in the process of obtaining the matching result of the to-be-searched text and the candidate commodity information according to all feature matching values, the third application server 303 can perform addition processing on all feature matching values ​​to obtain the matching result of the to-be-searched text and the candidate commodity information, or can perform weighted addition processing on all feature matching values ​​to obtain the matching result of the to-be-searched text and the candidate commodity information. Therefore, by adding commodity type tags to the commodity recall text recalled by the to-be-searched text according to the respective commodity text types before commodity matching, it is possible to distinguish information of different importance in the to-be-searched text, improve the difference of features during commodity matching, thereby improving the accuracy of commodity matching and improving the commodity chain pointing effect.

[0102] Since the commodity feature information corresponds to commodity type information, the commodity type information includes commodity key information, the commodity text type includes text key type, and different types of information have different importance to the overall text information (i.e., the text to be searched and the candidate commodity information), after the third application server 303 obtains the matching result of the text to be searched and the candidate commodity information, the third application server 303 can match the commodity feature information corresponding to the commodity key information with the commodity text feature information corresponding to the text key type; when the matching result is that the text to be searched and the candidate commodity information do not match, and the commodity feature information corresponding to the commodity key information is consistent with the commodity text feature information corresponding to the text key type, an error correction mark is added to the text to be searched and the candidate commodity information, and then, the error correction mark can be used to match the text to be searched and the candidate commodity information. Error mark, perform secondary matching processing on the text to be searched and the candidate product information to obtain the secondary matching result of the text to be searched and the candidate product information; in addition, when the matching result is that the text to be searched and the candidate product information do not match, and the product feature information corresponding to the product key information is consistent with the product text feature information corresponding to the text key type, add an error correction mark to the text to be searched and the candidate product information, and then, according to the error correction mark, perform secondary matching processing on the text to be searched and the candidate product information to obtain the secondary matching result of the text to be searched and the candidate product information. By verifying the matching result, the text to be searched corresponding to the matching result with potential errors and the candidate product information are matched secondary, which helps to improve the accuracy of product matching and improve the product chain pointing effect.

[0103] It should be noted that in various specific embodiments of the present invention, when it comes to the need to perform relevant processing based on data related to the characteristics of the target object such as the target object attribute information or attribute information set, the permission or consent of the target object will be obtained first, and the collection, use and processing of these data will comply with relevant laws, regulations and standards. Among them, the target object can be a user. In addition, when the embodiment of the present invention needs to obtain the attribute information of the target object, it will obtain the separate permission or separate consent of the target object through a pop-up window or jump to a confirmation page, and after clearly obtaining the separate permission or separate consent of the target object, the necessary target object-related data used to enable the normal operation of the embodiment of the present invention will be obtained.

[0104] Figure 5 is a flowchart of a commodity matching method provided by an embodiment of the present invention, and the commodity matching method can be executed by a server. Figure 5 The product matching method may include but is not limited to steps 510 to 550.

[0105] Step 510: Acquire the text to be searched, perform text recall processing on the text to be searched, and obtain multiple product recall texts, each of which corresponds to a product text type.

[0106] In one embodiment, the text to be searched may refer to text information used for product linking with candidate product information in the candidate product link library. The text to be searched may be product information, such as text information with words such as title, product word, brand word or product attribute, and the product information may be obtained by deleting text information from the candidate product link library, so that the product information can be used to detect whether the matching result of the product matching system in the server is correct, or the product information can be used to perform sample training on the product matching system. The product information can also be obtained by user upload or obtained from an open source information source. The text to be searched may also be a search term in a search application. For example, the search term entered in the search application of the user terminal can be obtained in real time, and the entered search term can be processed for product matching in real time, and the real-time matching result can be fed back to the user terminal. In addition, the text to be searched may also be network text information obtained from an open source information source, such as public account articles, news, blogs, etc. Among them, when it is necessary to use the search term in the search application or the network text information in the open source information as the text to be searched, it is necessary to perform semantic recognition processing on the search term or the network text information to determine whether the search term or the network text information contains product information (such as product words, brand words or product attributes, etc.); if the search term or the network text information contains product information, the search term or the network text information can be used as the text to be searched.

[0107] Reference Figure 6 , Figure 6 1 is a schematic diagram of a text recall process provided by an embodiment of the present invention. In one embodiment, the product recall text may refer to standard text information corresponding to or similar to the meaning expressed by the text to be searched. The product recall text is obtained by performing text recall processing on the text to be searched. Each product recall text corresponds to a product text type, for example, Figure 6As shown, the text to be searched is [Hyaluronic acid smooth hair care 100ml]. By performing text recall processing on the text to be searched, multiple product recall texts can be obtained, wherein one product recall text is [Hair shampoo and hair care], and the product text type is [category type]; another product recall text is [Net content 100 ml], and the product text type is [attribute type]. Since the text to be searched can be obtained from network text in product information or search terms or open source information sources, the text to be searched generally includes multiple types of texts, and different types of texts have different and highly different meanings. At the same time, since the information sources of the text to be searched are diverse, the text styles of the text to be searched are also different, and the text styles of the candidate product information in the candidate product link library are also different. In the related art, the text information to be matched (i.e., the text to be retrieved) is directly retrieved and matched from the product library, but the differences between different types of texts will be ignored, affecting the accuracy of the retrieval matching, and the direct matching of information with different text styles will also affect the accuracy of the retrieval matching. Therefore, by first performing text recall processing on the text to be retrieved, identifying and extracting different types of information in the search text, obtaining text information of different product text types, and improving the richness of the information to be matched, and at the same time unifying the text form of the text information, it helps to improve the accuracy of retrieval matching.

[0108] Reference Figure 7 , Figure 7 is a schematic diagram of a text recall process provided by another embodiment of the present invention. In one embodiment, the product recall text may include multiple product text words, and the product text types corresponding to the multiple product text words of the same product recall text are the same, for example, Figure 7 As shown, the text to be searched is [Hyaluronic Acid Smooth Hair Care 100ml]. By performing text recall processing on the text to be searched, multiple product recall texts can be obtained, one of which may be [Hair Conditioner, Hair Cream, Hair Conditioner]. The product text words contained in the product recall text are [Hair Conditioner], [Hair Cream] and [Hair Conditioner], respectively. The product text type corresponding to the product recall text is [Product Type], and the product text type corresponding to these three product text words is also [Product Type].

[0109] In one embodiment, in the process of text recall processing for the text to be searched, semantic feature recognition can be performed on the text to be searched first to obtain multiple semantic feature information, and then text recall processing can be performed according to each semantic feature information to obtain multiple commodity recall texts. Since the text styles in the text to be retrieved are diverse, the information content with the same meaning is expressed by different text words in different text styles, and the importance of words with different meanings in the text to be retrieved is also different. The higher the importance of a word, the easier it is to distinguish it from other commodity information using the word. Therefore, by performing semantic feature recognition processing on the text to be searched, the meaning to be expressed by the text to be searched can be identified and extracted, and words with different meanings in the text to be searched can be distinguished, and then words of different importance can be distinguished, so as to avoid directly matching the text to be searched and misinterpreting the meaning to be expressed in the text to be searched, resulting in poor linking effect. By using each semantic feature information after distinguishing the meaning to perform text recall processing separately to obtain multiple commodity recall texts, the importance of different words can be highlighted, and the meaning of a single word in the text to be searched can be avoided from being ignored due to the overall meaning of the text to be searched for text recall, which helps to improve the accuracy of the commodity recall text. Among them, each semantic feature information corresponds to at least one product recall text, that is, one semantic feature information can obtain at least one product recall text through text recall processing. For example, the semantic feature information is [hair care], and two product recall texts [conditioner] and [hair shampoo] can be obtained through text recall processing. The product text type of the product recall text [conditioner] is [product type], and the product text type of the product recall text [hair shampoo] is [category type]. That is, the same semantic feature information can obtain multiple product recall texts of different product text types through text recall processing, thereby increasing the number of product recall texts and the content richness of the product recall texts.

[0110] In one embodiment, in the process of performing text recall processing according to each semantic feature information, text recall processing can be performed according to each semantic feature information to obtain multiple candidate product recall texts, then, multiple supplementary product recall texts are obtained according to the candidate product recall texts and the candidate product information, and then, multiple product recall texts are obtained according to all candidate product recall texts and all supplementary product recall texts. Since the text form of the text to be searched is different from the text form of the candidate product information, the text to be searched may lack some feature information relative to the candidate product, resulting in the obtained product recall text being incomplete for the candidate product information, for example, relative to the candidate product information, some text types of product recall text may be missing, thereby affecting the accuracy of product matching. In order to improve the accuracy of product matching, multiple semantic feature information can be used to perform text recall processing to obtain multiple candidate product recall texts. Then, the candidate product recall texts are compared with the candidate product information to determine the missing supplementary product recall texts. The missing supplementary product recall texts are then integrated with the candidate product recall texts to obtain multiple product recall texts, thereby increasing the number of product recall texts and ensuring that the text to be searched contains text information to be matched corresponding to the candidate product information, thereby improving the accuracy of product matching.

[0111] In one embodiment, in the process of text recall processing of semantic feature information, the content of the product recall text can be restricted according to the product text type contained in the text to be searched, that is, the semantic feature information is only recalled within the scope of the corresponding product text type, so that accurate text recall can be achieved to avoid the semantics of the recalled product recall text deviating from the semantics of the text to be searched due to the excessive scope of text recall. In addition, in the process of text recall of semantic feature information, the product text type contained in the text to be searched can be supplemented and expanded to enrich the type of recall text, for example, the type in the candidate product information can be introduced, or the type related to the product text type contained can be introduced, so as to increase the number of product recall texts and enrich the content of product recall texts.

[0112] In one embodiment, in the process of recalling candidate commodity texts and candidate commodity information, the supplementary commodity text type corresponding to the candidate commodity recall text can be determined according to the candidate commodity recall text and the candidate commodity information, and then, multiple supplementary commodity recall texts can be obtained according to the supplementary commodity text type. Since each semantic feature information is obtained by distinguishing information of different commodity text types in the text to be searched, each semantic feature information has a corresponding commodity text type, and the candidate commodity information includes feature information corresponding to different commodity type information, and the text form of the text to be searched is different from the text form of the candidate commodity information, the text to be searched may lack feature information corresponding to some commodity type information relative to the candidate commodity, resulting in the lack of commodity recall text corresponding to some commodity type information. Therefore, by comparing the commodity text type in the candidate commodity recall text with the commodity type information of the candidate commodity information, the commodity type information missing in the candidate commodity recall text is determined, so that the missing commodity type information is determined as the supplementary commodity text type corresponding to the candidate commodity recall text, and then the semantic feature information is subjected to secondary text recall processing according to the supplementary commodity text type, and multiple supplementary commodity recall texts are obtained, thereby improving the richness of the commodity recall text and the accuracy of commodity matching.

[0113] Reference Figure 8 , Figure 8 FIG. 1 is a schematic diagram of a text recall process provided by another embodiment of the present invention. Figure 8As shown, the text to be searched is [Brand A Softening Hair Cream], and the candidate product recall texts obtained are [Brand A], [Softening] and [Conditioner], among which the product text type of the candidate product recall text [Brand A] is [Brand Type], the product text type of the candidate product recall text [Softening] is [Attribute Type], and the product text type of the candidate product recall text [Conditioner, Conditioner] is [Product Type]; and the candidate product information is [Brand A Hyaluronic Acid Softening Hair Conditioner], among which the candidate product information includes candidate product sub-information [Brand A] with product type information of [Brand Type], candidate product sub-information [Conditioner] with product type information of [Product Type], candidate product sub-information [Hyaluronic Acid] with product type information of [Attribute Type], candidate product sub-information [Softening] with product type information of [Attribute Auxiliary Type], and candidate product sub-information [Personal Care Cleansing, Hair Care, Shampoo and Conditioner] with product type information of [Category Type]. By comparing the types of the candidate product recall text and the candidate product information, it can be determined that the product text type in the candidate product recall text lacks [attribute auxiliary type] and [category type]. Therefore, the two missing types [attribute auxiliary type] and [category type] can be used as supplementary product text types. Then, the semantic feature information can be processed for secondary text recall based on the supplementary product text types [attribute auxiliary type] and [category type] to obtain the supplementary product recall text [glossy] with the product text type [attribute auxiliary type] and the supplementary product recall text [personal care cleaning, hair washing and care, shampoo and hair care, shampoo] with the product text type [category type]. As a result, the two supplementary product recall texts can be integrated with the three candidate product recall texts to obtain five product recall texts, and the number and type of these product recall texts correspond to the number and type of the candidate product sub-information.

[0114] Step 520: adding a product type label to each product recall text according to the product text type corresponding to each product recall text to obtain a labeled product recall text.

[0115] In one embodiment, each product recall text corresponds to a product text type. Therefore, a product type label can be added to each product recall text according to the product text type corresponding to each product recall text to obtain a marked product recall text, wherein the product type label corresponding to the marked product recall text corresponds to its own product text type, that is, the product text type of each marked product recall text can be distinguished by the product type label in the marked product recall text. Product recall texts of different product text types express different meanings and different importances. The use of product type labels can distinguish the importance of different product recall texts, highlight the differences between different marked product recall texts, and help improve the matching accuracy of the text to be searched and the candidate product information.

[0116] Step 530: concatenate all the marked product recall texts to obtain the product texts to be processed, and extract features from the product texts to be processed to obtain product text feature information of each marked product recall text.

[0117] In one embodiment, after obtaining the marked commodity recall text, all the marked commodity recall texts are spliced ​​to obtain the commodity text to be processed, and then the commodity text to be processed is subjected to feature extraction to obtain the commodity text feature information of each marked commodity recall text. The marked commodity recall texts of different commodity text types all contain useful revelation information. All the marked commodity recall texts are spliced ​​together to form a single commodity text to be processed, and the single commodity text to be processed is used as the input in the feature extraction model, which can comprehensively utilize the revelation information contained in the marked commodity recall texts of different commodity text types, enhance the input effect of the feature extraction model, and thus help improve the accuracy and feature extraction effect of the feature extraction model. If each marked commodity recall text is used as an input of the feature extraction model separately, the input dimension of the feature extraction model will be increased, the amount of calculation and the complexity of the model will be increased, the commodity matching efficiency will be reduced, and it will be difficult to meet the required business performance. However, through text splicing processing, the marked commodity recall texts of multiple different commodity text types are merged into an input vector of the feature extraction model, reducing the input dimension of the feature extraction model, which can help simplify the model structure and reduce the calculation and storage overhead. In addition, concatenating multiple labeled product recall texts to form the product text to be processed can help to explore the correlation and differences between labeled product recall texts of different product text types, help improve the learning and generalization capabilities of the feature extraction model, and reduce the occurrence of overfitting and underfitting problems.

[0118] In one embodiment, a feature extraction model can be used to perform feature extraction processing on the product text to be processed to obtain product text feature information of each marked product recall text. Since the marked product recall text contains a product type label, the product type label can be used to perform separate feature extraction on each marked product recall text to obtain the product text feature information corresponding to each marked product recall text. The product type label can also be used to perform feature extraction on multiple marked product recall texts at the same time, and the correlation between the multiple marked product recall texts can be mined to obtain the corresponding product text feature information. Therefore, the number of product text feature information can be the same as or different from the number of marked product recall texts, and each product text feature information can correspond to one or more marked product recall texts.

[0119] In one embodiment, the feature extraction process for the commodity text to be processed may be feature encoding process for the commodity text to be processed, and the commodity text to be processed is encoded by using an encoder, and the representation vector of the commodity recall text of each commodity type label can be output, that is, the commodity text feature information of each labeled commodity recall text. The encoder may be a model structure such as a BERT structure, a CNN structure, or an RNN structure, which is not specifically limited in the embodiment of the present invention.

[0120] In one embodiment, in the text to be searched and the candidate product information, different types of information have different importances. The more important the information is, the more it can represent the main feature information of the product information and the more it can be distinguished from other product information. For example, when performing linking, it is necessary to ensure that the information corresponding to [brand type] in the text to be searched is consistent with the information corresponding to [brand type] in the candidate product information; if the information corresponding to [brand type] of the two is inconsistent, it can be considered that the two are unrelated. At the same time, when using the feature extraction model to extract features, since the feature extraction model will read the information in the order of the input data, the feature extraction model gives priority to the top-ranked information, that is, the top-ranked information represents the dominant position in the input data. Therefore, it is necessary to distinguish the importance of information such as [brand type], [product type], [attribute type], [category type], etc. in the text to be searched and the candidate product information, so as to avoid ignoring the importance of information when performing feature extraction and affecting the matching accuracy. In order to highlight the importance of different types of information and improve the accuracy of product matching, it is necessary to sort all the labeled product recall texts according to the importance of the information when concatenating all the labeled product recall texts to obtain the product text to be processed, and then concatenate all the labeled product recall texts in turn according to the sorting results to obtain the product text to be processed. Fig. 9 , Fig. 9The process diagram of splicing all the marked product recall texts to form the product text to be processed is shown. The marked product recall texts include {[text title type]: Brand A hair conditioner}, {[brand type]: Brand A}, {[product type]: hair conditioner, hair conditioner}, {[category type]: personal care cleaning, hair washing and care, shampoo and hair care, shampoo} and {[attribute type]: hyaluronic acid}. All the marked product recall texts are sorted according to the importance of different types of information to obtain all the sorted marked product recall texts, and then all the sorted marked product recall texts are sequentially spliced ​​to form the product text to be processed. The product text to be processed can be {[brand type]: Brand A; [product type]: hair conditioner, hair conditioner; [text title type]: Brand A hair conditioner; [category type]: personal care cleaning, hair washing and care, shampoo and hair care, shampoo; [attribute type]: hyaluronic acid}. Among them, the text to be searched can be directly used as one of the product recall texts, and the corresponding product text type is the text title type.

[0121] Step 540: Obtain candidate product information preset in a candidate product link library according to the text to be searched, and obtain multiple product feature information of the candidate product information.

[0122] In one embodiment, after obtaining the text to be searched, the candidate product information can be obtained from a preset candidate product link library according to the text to be searched, and multiple product feature information corresponding to the candidate product information can be obtained. Among them, the candidate product link library includes multiple candidate product information. In order to reduce the amount of data processing and improve the efficiency of product matching, the relevant or similar candidate product information can be searched from the candidate product link library according to the text to be searched, and the relevant or similar candidate product information can be preliminarily screened for product matching. For example, the relevant or similar candidate product information can be recalled from the candidate product link library according to the text to be searched. The preset candidate product link library can also store multiple product feature information of the candidate product information, wherein the product feature information can refer to the feature information corresponding to the text information of different types or different meanings in the candidate product information. Therefore, the pre-stored product offline data and the product text feature information obtained by online processing can be used to perform product matching processing, which can reduce the amount of data processed online, shorten the time of online processing, and improve the efficiency of product matching.

[0123] In one embodiment, the candidate product information includes a plurality of candidate product sub-information, and the candidate product sub-information corresponds to product type information, wherein the product type information of each candidate product sub-information may be different, and the meaning and importance expressed by the candidate product sub-information of different product type information are different. Therefore, a candidate product type label may be added to each candidate product sub-information according to the product type information to obtain the marked product information, and then all the marked product information are spliced ​​to obtain the product information to be processed. Among them, the candidate product type label corresponds to the product type information, so that the product type information of each marked product information can be distinguished by the candidate product type label in the marked product information, and at the same time, the importance of different candidate product sub-information can be distinguished, and the difference between different candidate product sub-information can be highlighted, which is helpful to improve the matching accuracy of the text to be searched and the candidate product information. In addition, the candidate product type label and the product type label can be used to distinguish the product recall text and the candidate product sub-information, which is helpful to distinguish different feature information when matching each feature information, and reduce the occurrence of the situation that the matching result is affected by data confusion. Different annotated product information contains different meaning information. All the annotated product information is spliced ​​and merged to form the product information to be processed. The product information to be processed is used as the input of the feature extraction model. The different meaning information contained in the annotated product information of different product types can be comprehensively utilized to enhance the input effect of the feature extraction model, thereby helping to improve the accuracy and feature extraction effect of the feature extraction model. Through information splicing processing, the annotated product information of multiple different product types is merged into an input vector of the feature extraction model, reducing the input dimension of the feature extraction model, which can help simplify the model structure and reduce the calculation and storage overhead. In addition, the information splicing processing of multiple annotated product information to form the product information to be processed can help to explore the correlation and difference between the annotated product information of different product types, help to improve the learning ability and generalization ability of the feature extraction model, and reduce the occurrence of overfitting and underfitting problems.

[0124] In one embodiment, a feature extraction model can be used to perform feature extraction processing on the product information to be processed to obtain product feature information of each labeled product information. Since the labeled product information contains candidate product type labels, the candidate product type labels can be used to perform separate feature extraction on each labeled product information to obtain the product feature information corresponding to each labeled product information. The candidate product type labels can also be used to perform feature extraction on multiple labeled product information at the same time, and the correlation between multiple labeled product information can be mined to obtain the corresponding product feature information. Therefore, the number of product feature information can be the same as or different from the number of labeled product information, and each product feature information can correspond to one or more labeled product information.

[0125] Reference Fig.10 , Fig.10 It is a schematic diagram of the process of obtaining product feature information provided by one embodiment of the present invention. In one embodiment, a preset candidate product link library includes multiple candidate product information. By preprocessing each candidate product information, the product feature information of each candidate product information is obtained. Among them, the preprocessing process includes the following: first, adding a candidate product type label to the candidate sub-information in the candidate product information to obtain the marked product information, and then splicing all the marked product information to obtain the product information to be processed, and then performing feature extraction processing on the product information to be processed, so as to obtain the product feature information of the candidate product information. After obtaining the product feature information of each candidate product information, the product feature information can be bound to the candidate product information, and then stored in the preset candidate product link library, or all product feature information can be stored in an additional feature center database.

[0126] In one embodiment, the feature extraction process for the commodity information to be processed may be feature encoding process for the commodity information to be processed, and the commodity text to be processed is encoded by using an encoder, and a representation vector of the candidate commodity sub-information of each candidate commodity type label may be output, that is, the commodity feature information of each labeled commodity information. Among them, the encoder may be a model structure such as a BERT structure, a CNN structure, or an RNN structure, and the embodiment of the present invention does not specifically limit this. In addition, the encoder used when performing feature encoding process on the commodity information to be processed may be the encoder used when performing feature encoding process on the commodity text to be processed, that is, the same encoder may be used to perform feature encoding process on the commodity information to be processed and the commodity text to be processed respectively; or, the encoder used when performing feature encoding process on the commodity information to be processed may be the same as or different from the encoder used when performing feature encoding process on the commodity text to be processed.

[0127] Step 550: Perform text matching processing on all product text feature information and all product feature information to obtain matching results between the text to be searched and the candidate product information.

[0128] In one embodiment, after obtaining the commodity text feature information and the commodity feature information, text matching processing can be performed on all commodity text feature information and all commodity feature information to obtain the matching result between the text to be searched and the candidate commodity information. Since the candidate commodity information and the related commodity feature information are obtained and deployed in advance by independent model calculation, after obtaining the text to be searched, only the text to be searched needs to be processed to obtain the commodity text feature information corresponding to the text to be searched. Therefore, after obtaining the pre-processed commodity feature information, real-time text matching processing can be performed on each commodity text feature information and each commodity feature information to determine the matching degree between the text to be searched and the current candidate commodity information, that is, to obtain the matching result between the text to be searched and the candidate commodity information. By adopting the method of post-interaction of all types of commodity feature information and commodity text feature information, the amount of real-time data processing can be reduced, the data processing efficiency can be improved, the accuracy of commodity matching can be improved while meeting business performance, and the chain finger effect can be improved. In addition, since each product text feature information corresponds to each marked product recall text, that is, each product text feature can have a corresponding product type label, and each product feature information corresponds to each marked product information, that is, each product feature information can also have a corresponding candidate product type label, therefore, each product text feature information and each product feature information can be distinguished by the product type label and the candidate product type label, which helps to match each product text feature information and each product feature information, improve the accuracy of the data during the matching process, and reduce the occurrence of data confusion.

[0129] In one embodiment, in the process of performing text matching processing on all commodity text feature information and all commodity feature information, each commodity text feature information can be subjected to text matching processing with all commodity feature information to obtain feature matching values ​​corresponding to each commodity text feature information, and then the matching result of the to-be-searched text and the candidate commodity information can be obtained according to all feature matching values. By performing text matching processing on each commodity text feature information and each commodity feature information respectively to obtain feature matching values ​​of each commodity text feature information and each commodity feature information, the correlation between each commodity text feature information and each commodity feature information can be mined, which is helpful to improve the accuracy of commodity matching.

[0130] In one embodiment, in the process of text matching each commodity text feature information with all commodity feature information, each commodity text feature information can be matched with each commodity feature information to obtain the feature similarity between each commodity text feature information and each commodity feature information, and then for each commodity text feature information, the feature similarity with the largest value is used as the feature matching value corresponding to each commodity text feature information. Since the commodity text feature information and the commodity feature information can be the representation vectors of the commodity recall text and the candidate commodity sub-information in each type, the cosine similarity can be calculated for each commodity text feature information and each commodity feature information, and the feature similarity with the largest value is used as the feature matching value corresponding to the commodity text feature information, that is, the similarity between the to-be-searched text and the candidate commodity information in the commodity text type corresponding to the commodity recall text, and then the similarity between the to-be-searched text and the candidate commodity information in each commodity text type can be calculated, which helps to improve the accuracy of the matching results.

[0131] Reference Fig.11 , Fig.11 Schematic diagram of the process of obtaining the matching result between the search text and the candidate product information provided by the embodiment of the present invention. Fig.11 As shown, after obtaining the commodity text to be processed, the commodity text to be processed can be input into the first encoder for feature extraction processing to obtain commodity text feature information of each commodity text type. After obtaining the commodity information to be processed, the commodity information to be processed can be input into the second encoder for feature extraction processing to obtain commodity feature information of each commodity type information, wherein the feature extraction processing performed on the commodity information to be processed is performed before obtaining the commodity text to be processed, and the structures and parameters of the first encoder and the second encoder can be the same or different. After obtaining the commodity text feature information and the commodity feature information, similarity matching is performed on all commodity feature information according to each commodity text feature information, and the feature similarity with the largest value is used as the feature matching value corresponding to the commodity text feature information, thereby obtaining the feature matching value of each commodity text feature information. Then, the matching result of the text to be searched and the candidate commodity information is determined according to the feature matching value of each commodity text feature information, wherein the matching result can be obtained by adding or weighted adding the feature matching values ​​of each commodity text feature information.

[0132] In one embodiment, in the process of obtaining the matching result of the text to be searched and the candidate product information according to all feature matching values, all feature matching values ​​can be added to obtain the matching result of the text to be searched and the candidate product information; or, all feature matching values ​​can be weighted added to obtain the matching result of the text to be searched and the candidate product information. The overall matching degree between the text to be searched and the candidate product information is obtained by adding or weighted adding the matching degree of the text to be searched and the candidate product information in each product text type. Among them, since the importance of text information corresponding to different product text types is different, weighting coefficients can be set according to different product text types. For example, for product information, brand information and product information are both of high importance, while the importance of text information such as category information and attribute information is relatively low. Therefore, different weighting coefficients can be set according to the importance corresponding to different types, which can highlight the importance of different text information and improve the matching accuracy.

[0133] In one embodiment, there may be multiple candidate product information. In the process of text matching all product text feature information and all product feature information, text matching can be performed on all product text feature information and the product feature information corresponding to each candidate product information to obtain candidate matching results between the text to be searched and each candidate product information. Then, according to the size of each candidate matching result, each candidate product information is sorted to obtain matching results between the text to be searched and the candidate product information. Since the candidate product link library includes multiple product information, in order to reduce the data processing amount of product matching, product information related to or similar to the text to be searched is obtained from the candidate product link library as candidate product information. There may be multiple product information related to or similar to the text to be searched, and multiple related or similar product information are all used as candidate product information, so that multiple candidate product information can be obtained, thereby improving the product chain pointing effect. By matching the text to be searched with each candidate product information separately, the candidate matching results of the text to be searched and each candidate product information are obtained; wherein, the candidate matching results of the text to be searched and each candidate product information can be represented by a numerical value, that is, the size of the numerical value is used to represent the matching degree between the text to be searched and the candidate product information. If the numerical value is larger, the matching degree between the text to be searched and the candidate product information is higher; if the numerical value is smaller, the matching degree between the text to be searched and the candidate product information is lower. Therefore, the candidate matching results corresponding to each candidate product information can be used to sort each candidate product information from large to small to obtain the matching results of the text to be searched and the candidate product information, thereby improving the accuracy of product matching and improving the product chain pointing effect.

[0134] Reference Fig.12 , Fig.12The present invention provides a schematic diagram of a process of obtaining a matching result between a text to be searched and multiple candidate product information provided by an embodiment of the present invention. In one embodiment, the obtained candidate product information has multiple items. All product text feature information of the text to be searched and the product feature information corresponding to each candidate product information can be respectively processed for similarity matching to obtain a candidate matching result between the text to be searched and each candidate product information, that is, the similarity between the text to be searched and each candidate product information. The higher the similarity between the text to be searched and the candidate product information, the higher the correlation between the text to be searched and the candidate product information. Therefore, all candidate product information can be sorted according to the numerical value of each candidate matching result to form a product sorting list, that is, the matching result between the text to be searched and the candidate product information. Among them, the product sorting list can include each candidate product information and the corresponding candidate matching result. In addition, when the candidate matching result is that the text to be searched does not match the candidate product information (such as the similarity between the text to be searched and the candidate product information is 0), the candidate product information can not participate in the sorting, that is, the matching result does not include the candidate product information.

[0135] In one embodiment, the candidate product information corresponding to the candidate matching result with the largest value may be used as the matching result between the text to be searched and the candidate product information, or multiple candidate product information may be selected according to the sorting result as the matching result between the text to be searched and the candidate product information, or a sorted list may be generated according to the sorted candidate product information, and the sorted list may be used as the matching result between the text to be searched and the candidate product information.

[0136] In one embodiment, the commodity feature information corresponds to commodity type information, the commodity type information includes commodity key information, and the commodity text type includes text key type. After obtaining the matching result of the text to be searched and the candidate commodity information, it can be determined whether the candidate commodity information meets the link requirement of the text to be searched according to the matching result, commodity key information and text key type. When the matching result is that the text to be searched does not match the candidate commodity information, and the commodity feature information corresponding to the commodity key information is consistent with the commodity text feature information corresponding to the text key type, an error correction mark is added to the text to be searched and the candidate commodity information, and then, according to the error correction mark, the text to be searched and the candidate commodity information are subjected to secondary matching processing to obtain the secondary matching result of the text to be searched and the candidate commodity information. The commodity type information of different commodity feature information is different, and the commodity feature information corresponding to different commodity type information has different meanings and different importance. The more important the information is, the more it can represent the main feature information of the candidate commodity information. Therefore, the commodity type information with high importance can be the commodity key information, for example, the commodity key information can be the brand type and product type. Correspondingly, the text key type can be the commodity text type with high importance, and the text key type can be the same as the commodity key information. If the commodity feature information corresponding to the commodity key information is consistent with the commodity text feature information corresponding to the text key type, then the candidate commodity information can be considered to be related or similar to the text to be searched. For example, the text key attribute is determined based on the commodity key information, the text key attribute is the same as the commodity key information, and the text key attribute and the commodity key information both include product type and brand type. When the commodity feature information corresponding to the product type is consistent with the commodity text feature information corresponding to the product type, and the commodity feature information corresponding to the brand type is consistent with the commodity text feature information corresponding to the brand type, then the candidate commodity information can be considered to be related or similar to the text to be searched. Therefore, when the matching result is that the text to be searched and the commodity to be searched are not matched, and the commodity feature information corresponding to the commodity key information is consistent with the commodity text feature information corresponding to the text key type, then it can be considered that there is a potential error in the matching result, thereby adding corresponding error correction marks to the text to be searched and the candidate commodity information to reduce the occurrence of chain finger errors. Then, the text to be searched with the corresponding error correction mark and the candidate commodity information can be re-matched, that is, a secondary matching process can be performed to re-judge whether the text to be searched and the candidate commodity information match, thereby reducing the chain finger error rate.

[0137] In one embodiment, the commodity feature information corresponds to commodity type information, the commodity type information includes commodity key information, and the commodity text type includes text key type. After obtaining the matching result of the text to be searched and the candidate commodity information, it can be determined whether the candidate commodity information meets the link requirement of the text to be searched based on the matching result, commodity key information and text key type. When the matching result is that the text to be searched matches the candidate commodity information, and the commodity feature information corresponding to the commodity key information is inconsistent with the commodity text feature information corresponding to the text key type, an error correction mark is added to the text to be searched and the candidate commodity information, and then, according to the error correction mark, the text to be searched and the candidate commodity information are subjected to secondary matching processing to obtain the secondary matching result of the text to be searched and the candidate commodity information. The commodity type information of different commodity feature information is different, and the commodity feature information corresponding to different commodity type information has different meanings and different importance. The more important the information is, the more it can represent the main feature information of the candidate commodity information. Therefore, the commodity type information with high importance can be the commodity key information, for example, the commodity key information can be the brand type and product type. Correspondingly, the text key type can be a commodity text type with high importance, and the text key type can be the same as the commodity key information. For example, the text key attribute is determined based on the commodity key information, and the text key attribute is the same as the commodity key information. The text key attribute and the commodity key information both include attribute type and brand type. When the commodity feature information corresponding to the attribute type is inconsistent with the commodity text feature information corresponding to the attribute type, and the commodity feature information corresponding to the brand type is inconsistent with the commodity text feature information corresponding to the brand type, then it can be considered that the candidate commodity information is irrelevant or dissimilar to the text to be searched. If the commodity feature information corresponding to the commodity key information is inconsistent with the commodity text feature information corresponding to the text key type, then it can be considered that the candidate commodity information is irrelevant or dissimilar to the text to be searched. Therefore, when the matching result is that the text to be searched matches the commodity to be searched, and the commodity feature information corresponding to the commodity key information is inconsistent with the commodity text feature information corresponding to the text key type, then it can be considered that there is a potential error in the matching result, so that corresponding error correction marks are added to the text to be searched and the candidate commodity information to reduce the occurrence of link errors. Next, the text to be searched with the corresponding error correction mark and the candidate product information may be re-matched, that is, a secondary matching process may be performed to re-determine whether the text to be searched and the candidate product information match, thereby reducing the link error rate.

[0138] Reference Fig.13 , Fig.13The process diagram of the commodity matching method provided by the embodiment of the present invention is as follows. In one embodiment, a candidate commodity link library is pre-set, and the candidate commodity link library includes multiple candidate commodity information. By pre-performing offline model calculation on each candidate commodity information, that is, model calculation without interacting with the text to be searched, the commodity feature information of each candidate commodity information is obtained, and the commodity feature information is stored in the feature database to complete the deployment of offline data. When a search request initiated by a user terminal is received, the text to be searched in the search request is obtained, and the model calculation is performed in real time on the text to be searched to obtain the commodity text feature information of the text to be searched. At the same time, the candidate commodity link library preset in the text to be searched is used to obtain the relevant candidate commodity information, and the commodity feature information corresponding to the candidate commodity information is obtained from the feature database. After obtaining the pre-deployed commodity feature information and the commodity text feature information obtained by real-time calculation, the commodity feature information and the commodity text feature information are matched and processed to obtain the matching result of the text to be searched and the candidate commodity information. Since the candidate product information and related product feature information are obtained and deployed in advance through independent model calculations, after obtaining the text to be searched, only the text to be searched needs to be processed to obtain the product text feature information corresponding to the text to be searched. Therefore, by adopting a post-interaction method for all types of product feature information and product text feature information, the real-time data processing volume can be reduced, the data processing efficiency can be improved, the accuracy of product matching can be improved while meeting business performance, and the chain finger effect can be improved.

[0139] The product matching method provided by the embodiment of the present invention is described in detail below with reference to a specific example.

[0140] Reference Fig.14 As shown, Fig.14 This is a flowchart of the steps of the product matching method provided by a specific example. Fig.14 In the example, the product matching method may include but is not limited to steps 1401 to 1423.

[0141] Step 1401: Get the text to be searched.

[0142] Step 1402: Perform semantic feature recognition on the text to be searched to obtain multiple semantic feature information.

[0143] In this step, by performing semantic feature recognition processing on the search text, the meaning that the search text needs to express can be identified and extracted, and words with different meanings in the search text can be distinguished, and then words of different importance can be distinguished, avoiding direct matching of the search text and misinterpreting the meaning that needs to be expressed in the search text, resulting in poor chaining effect.

[0144] Step 1403: Perform text recall processing according to each semantic feature information to obtain multiple candidate product recall texts.

[0145] In this step, the semantic feature information after the meaning is distinguished is used to perform text recall processing respectively, and multiple candidate product recall texts are obtained, which can highlight the importance of different words and avoid ignoring the meaning of individual words in the text to be searched due to the use of the overall meaning of the text to be searched for text recall, which helps to improve the accuracy of the candidate product recall text. Among them, each semantic feature information corresponds to at least one candidate product recall text, that is, one semantic feature information can obtain at least one candidate product recall text through text recall processing.

[0146] Step 1404: Determine the type of supplementary product text corresponding to the candidate product recall text according to the candidate product recall text and the candidate product information.

[0147] Step 1405: Acquire multiple supplementary product recall texts according to the supplementary product text type.

[0148] Step 1406: Obtain multiple product recall texts based on all candidate product recall texts and all supplementary product recall texts.

[0149] In this step, by comparing the product text type in the candidate product recall text with the product type information of the candidate product information, the product type information missing from the candidate product recall text is determined, and the missing product type information is determined as the supplementary product text type corresponding to the candidate product recall text, and then according to the supplementary product text type, a secondary text recall processing is performed on the semantic feature information to obtain multiple supplementary product recall texts, thereby improving the richness of the product recall text and improving the accuracy of product matching.

[0150] Step 1407: adding a product type label to each product recall text according to the product text type corresponding to each product recall text to obtain a labeled product recall text.

[0151] Step 1408: concatenate all the marked product recall texts to obtain the product text to be processed.

[0152] Step 1409: extracting features from the product text to be processed to obtain product text feature information of each labeled product recall text.

[0153] Step 1410: Obtain candidate product information preset in a candidate product link library according to the text to be searched.

[0154] In this step, the pre-set candidate product link library includes multiple candidate product information. The pre-stored product offline data and the product text feature information obtained by online processing can be used to perform product matching processing, which can reduce the amount of data processed online, shorten the time for online processing, and improve the efficiency of product matching.

[0155] Step 1411: Acquire multiple commodity feature information of candidate commodity information.

[0156] In this step, the candidate product information includes multiple candidate product sub-information, and the candidate product sub-information corresponds to product type information. The product type information of each candidate product sub-information can be different, and the meaning and importance expressed by the candidate product sub-information of different product type information are different. Therefore, a candidate product type label can be added to each candidate product sub-information according to the product type information to obtain the marked product information, and then all the marked product information is spliced ​​to obtain the product information to be processed. In addition, the candidate product type label and the product type label can be used to distinguish the product recall text and the candidate product sub-information, which helps to distinguish different feature information when matching each feature information, and reduce the occurrence of situations where the matching results are affected by data confusion.

[0157] Step 1412: Determine the number of candidate product information obtained. If there is one candidate product information, execute step 1413; if there are multiple candidate product information, execute step 1413 or step 1416.

[0158] Step 1413: Perform similarity matching processing on each commodity text feature information and each commodity feature information to obtain feature similarity between each commodity text feature information and each commodity feature information.

[0159] Step 1414: For each product text feature information, the feature similarity with the largest corresponding value is used as the feature matching value corresponding to each product text feature information.

[0160] Step 1415 : perform addition or weighted addition processing on all feature matching values ​​to obtain the matching result between the text to be searched and the candidate product information, and execute step 1418 .

[0161] Step 1416: Perform text matching processing on all product text feature information and product feature information corresponding to each candidate product information to obtain candidate matching results between the text to be searched and each candidate product information.

[0162] Step 1417: Sort each candidate product information according to the size of each candidate matching result to obtain the matching result between the text to be searched and the candidate product information.

[0163] Step 1418: Determine the matching result. If the matching result is a match, execute step 1419; if the matching result is inconsistent, execute step 1420.

[0164] Step 1419: Determine whether the product feature information corresponding to the product key information is inconsistent with the product text feature information corresponding to the text key type. If so, execute step 1421; if not, execute step 1423.

[0165] Step 1420: Determine whether the product feature information corresponding to the product key information is consistent with the product text feature information corresponding to the text key type. If so, execute step 1421; if not, execute step 1423.

[0166] Step 1421: Add error correction marks to the text to be searched and the candidate product information.

[0167] Step 1422: According to the error correction mark, the text to be searched and the candidate product information are subjected to secondary matching processing to obtain the secondary matching result of the text to be searched and the candidate product information.

[0168] Step 1423: End the step process.

[0169] Through the commodity matching method of the above steps 1401 to 1423, after obtaining the text to be searched, the text to be searched is subjected to text recall processing to obtain multiple commodity recall texts, each of which has a corresponding commodity text type, so that text information of different commodity text types can be obtained, thereby improving the richness of the information to be matched. Then, the commodity text type of each commodity recall text can be used to add a commodity type label to each commodity recall text to obtain a marked commodity recall text, thereby distinguishing text information of different commodity text types. Since the text content of different commodity text types has different importance for commodity information, by distinguishing information of different importance in the text to be searched, the difference of each feature in the text to be searched is highlighted, which helps to improve the matching accuracy of the text to be searched and the candidate commodity information. Then, the commodity text to be processed can be obtained by text splicing all the marked commodity recall texts, and feature extraction can be performed to obtain the commodity text feature information of each marked commodity recall text, so that the obtained commodity text feature information can retain the difference between different marked commodity recall texts, thereby improving the matching accuracy. By obtaining preset candidate product information and corresponding multiple product feature information based on the text to be searched, relevant candidate product information can be preliminarily screened for product matching, and at the same time, the amount of data processing for the candidate product information can be reduced, which helps to improve the accuracy and efficiency of product matching. By performing text matching processing on the product text feature information and the product feature information, the matching result between the text to be searched and the candidate product information can be determined, and then the product information associated with the text to be searched can be determined. Therefore, by adding product type tags to the product recall text recalled by the text to be searched before product matching according to the respective product text types, it is possible to distinguish information of different importance in the text to be searched, improve the difference in features during product matching, thereby improving the accuracy of product matching and improving the product chain pointing effect.

[0170] It is to be understood that, although the steps in the above-mentioned various flow charts are shown in sequence according to the indication of the arrows, these steps are not necessarily performed in sequence according to the order indicated by the arrows. Unless there is a clear explanation in the present embodiment, the execution of these steps does not have a strict order restriction, and these steps can be performed in other orders. Moreover, at least a portion of the steps in the above-mentioned flow charts can include a plurality of steps or a plurality of stages, and these steps or stages are not necessarily performed at the same time, but can be performed at different times, and the execution order of these steps or stages is not necessarily performed in sequence, but can be performed in turn or alternately with at least a portion of the steps or stages in other steps or other steps.

[0171] The following are some practical examples to illustrate the application scenarios of the embodiments of the present invention.

[0172] It should be noted that the product matching method provided in the embodiments of the present invention can be applied to different application scenarios such as generation and update of product display data during the development of a shopping platform, retrieval and query of product data during the operation of a shopping platform, retrieval and query of product data during the operation of a search engine system, and push of product data in a multimedia platform. The following takes the scenario of retrieval and query of product data during the operation of a search engine system and the scenario of push of product data in a multimedia platform as examples for explanation.

[0173] Scene 1

[0174] The commodity matching method provided in the embodiment of the present invention can be applied to the scene of searching and querying commodity data during the operation of the search engine system. For example, assuming that the application server is a cloud server running the search engine system and deploying a commodity matching module, and the application server stores a candidate commodity link library, the candidate commodity link library includes a plurality of candidate commodity information. After the user inputs the text to be searched through the search engine application in the user terminal (such as a mobile phone) that can be used to access the search engine system, the user terminal can transmit the text to be searched to the application server, so that the application server matches the commodity to the text to be searched, thereby achieving the purpose of linking. After the application server obtains the text to be searched, the application server can perform semantic feature recognition on the text to be searched to obtain multiple semantic feature information, and then perform text recall processing based on each semantic feature information to obtain multiple candidate product recall texts; after obtaining the candidate product recall text, the supplementary product text type corresponding to the candidate product recall text can be determined based on the candidate product recall text and the candidate product information, and then multiple supplementary product recall texts can be obtained based on the supplementary product text type, and then multiple product recall texts can be obtained based on all candidate product recall texts and all supplementary product recall texts; after obtaining multiple product recall texts, a product type label can be added to each product recall text based on the product text type corresponding to each product recall text to obtain annotated product recall text, and then all annotated product recall texts can be concatenated to obtain the product text to be processed, and then feature extraction can be performed on the product text to be processed to obtain product text feature information of each annotated product recall text; in addition, after the application server obtains the text to be searched, the application server can also According to the text to be searched, candidate product information preset in the candidate product link library is obtained, and multiple product feature information of the candidate product information is obtained. Then, each product text feature information can be matched with each product feature information to obtain the feature similarity between each product text feature information and each product feature information. Then, for each product text feature information, the feature similarity with the largest corresponding value is used as the feature matching value corresponding to each product text feature information. After obtaining the feature matching values ​​corresponding to all product text feature information, all feature matching values ​​can be added or weighted added to obtain the matching result of the text to be searched and the candidate product information. In addition, in the case that there are multiple candidate product information, the application server can perform text matching processing on all product text feature information and the product feature information corresponding to each candidate product information to obtain the candidate matching result of the text to be searched and each candidate product information. Then, according to the size of each candidate matching result, each candidate product information is sorted to obtain the matching result of the text to be searched and the candidate product information.After obtaining the matching result, the matching result, the product feature information corresponding to the product key information, and the product text feature information corresponding to the text key type can be judged. When the matching result is that the text to be searched does not match the candidate product information, and the product feature information corresponding to the product key information is consistent with the product text feature information corresponding to the text key type, an error correction mark is added to the text to be searched and the candidate product information; and when the matching result is that the text to be searched does not match the candidate product information, and the product feature information corresponding to the product key information is consistent with the product text feature information corresponding to the text key type, an error correction mark is added to the text to be searched and the candidate product information; then the application server can perform secondary matching processing on the text to be searched and the candidate product information according to the error correction mark to obtain a secondary matching result of the text to be searched and the candidate product information. Therefore, by using the text to be searched inputted in the user terminal, model calculation processing such as text recall, label addition, feature extraction, etc. is performed in real time to obtain product text feature information, and then the product feature information of the candidate product information obtained in advance through model calculation processing is used to perform feature matching processing, so that it can be determined whether the candidate product information matches the text to be searched. If they match, the candidate product information can be displayed on the user terminal. If they do not match, the candidate product information can be re-acquired and matched again. ;

[0175] Scene 2

[0176] The product matching method provided in the embodiment of the present invention can be applied to the scenario of pushing product data in a multimedia platform. For example, it is assumed that the application server is a cloud server that runs a multimedia platform and is deployed with a product matching module. At the same time, the application server stores a candidate product link library, which includes multiple candidate product information. The user initiates a request to access a network article in a multimedia platform to the application server through a user terminal (such as a computer). The application server can obtain the network article in the multimedia platform according to the access request, and use the network article as the text to be searched. Then, semantic feature recognition can be performed on the text to be searched to obtain multiple semantic feature information, and then text recall processing can be performed based on each semantic feature information to obtain multiple candidate product recall texts; after obtaining the candidate product recall text, the type of supplementary product text corresponding to the candidate product recall text can be determined based on the candidate product recall text and the candidate product information, and then, multiple supplementary product recall texts can be obtained based on the supplementary product text type, and then, multiple product recall texts can be obtained based on all candidate product recall texts and all supplementary product recall texts; after obtaining multiple product recall texts, a product type label can be added to each product recall text according to the product text type corresponding to each product recall text to obtain a marked product recall text, and then all marked product recall texts can be concatenated to obtain the product text to be processed, and then feature extraction can be performed on the product text to be processed to obtain the product text of each marked product recall text feature information; in addition, after the application server obtains the text to be searched, the application server can also obtain the candidate product information preset in the candidate product link library according to the text to be searched, and obtain multiple product feature information of the candidate product information, and then each product text feature information can be matched with each product feature information to obtain the feature similarity between each product text feature information and each product feature information, and then for each product text feature information, the feature similarity with the largest corresponding value is used as the feature matching value corresponding to each product text feature information; after obtaining the feature matching values ​​corresponding to all product text feature information, all feature matching values ​​can be added or weighted added to obtain the matching result of the text to be searched and the candidate product information; in addition, in the case that there are multiple candidate product information, the application server can perform text matching processing on all product text feature information and the product feature information corresponding to each candidate product information to obtain the candidate matching result of the text to be searched and each candidate product information, and then sort each candidate product information according to the size of each candidate matching result to obtain the matching result of the text to be searched and the candidate product information;After obtaining the matching result, the matching result, the product feature information corresponding to the product key information, and the product text feature information corresponding to the text key type can be judged. When the matching result is that the text to be searched does not match the candidate product information, and the product feature information corresponding to the product key information is consistent with the product text feature information corresponding to the text key type, an error correction mark is added to the text to be searched and the candidate product information; and when the matching result is that the text to be searched does not match the candidate product information, and the product feature information corresponding to the product key information is consistent with the product text feature information corresponding to the text key type, an error correction mark is added to the text to be searched and the candidate product information; then the application server can perform secondary matching processing on the text to be searched and the candidate product information according to the error correction mark to obtain a secondary matching result of the text to be searched and the candidate product information. Therefore, by using the text to be searched initiated in the user terminal, model calculation processing such as text recall, tag addition, feature extraction, etc. is performed in real time to obtain product text feature information, and then the product feature information of the candidate product information obtained in advance through model calculation processing is used to perform feature matching processing, so that it can be determined whether the candidate product information matches the text to be searched. If they match, the candidate product information and the network article can be displayed on the user terminal to realize the advertising push of product data. If they do not match, the candidate product information can be re-acquired and matched again. ;

[0177] Reference Fig.15 The embodiment of the present invention further discloses a product matching device. The product matching device 1500 can implement the product matching method in the previous embodiment. The product matching device 1500 includes:

[0178] The text recall unit 1501 is used to obtain the text to be searched, perform text recall processing on the text to be searched, and obtain multiple product recall texts, each of which corresponds to a product text type;

[0179] The label adding unit 1502 is used to add a product type label to each product recall text according to the product text type corresponding to each product recall text, so as to obtain a marked product recall text;

[0180] The feature extraction unit 1503 is used to perform text splicing on all the marked product recall texts to obtain the product texts to be processed, and perform feature extraction on the product texts to be processed to obtain product text feature information of each marked product recall text;

[0181] The information acquisition unit 1504 is used to acquire candidate product information preset in the candidate product link library according to the text to be searched, and acquire multiple product feature information of the candidate product information;

[0182] The text matching unit 1505 is used to perform text matching processing on all commodity text feature information and all commodity feature information to obtain matching results between the text to be searched and the candidate commodity information.

[0183] In one embodiment, the text recall unit 1501 is further used to:

[0184] Perform semantic feature recognition on the search text to obtain multiple semantic feature information;

[0185] The text recall process is performed according to each semantic feature information to obtain multiple product recall texts.

[0186] In one embodiment, the text recall unit 1501 is further used to:

[0187] Perform text recall processing according to each semantic feature information to obtain multiple candidate product recall texts;

[0188] Obtaining multiple supplementary product recall texts according to the candidate product recall text and the candidate product information;

[0189] According to all candidate product recall texts and all supplementary product recall texts, multiple product recall texts are obtained.

[0190] In one embodiment, the text recall unit 1501 is further used to:

[0191] Determine the type of supplementary product text corresponding to the candidate product recall text according to the candidate product recall text and the candidate product information;

[0192] Get multiple supplementary product recall texts based on the supplementary product text type.

[0193] In one embodiment, the information acquisition unit 1504 is further configured to:

[0194] According to the product type information, a candidate product type label is added to each candidate product sub-information to obtain the labeled product information;

[0195] All the marked product information is concatenated to obtain the product information to be processed, and the feature extraction of the product information to be processed is performed to obtain the product feature information of each marked product information;

[0196] The candidate product information includes a plurality of candidate product sub-information, and the candidate product sub-information corresponds to product type information.

[0197] In one embodiment, the text matching unit 1505 is further used to:

[0198] Perform text matching processing on each commodity text feature information and all commodity feature information to obtain feature matching values ​​corresponding to each commodity text feature information;

[0199] According to all feature matching values, the matching results of the search text and the candidate product information are obtained.

[0200] In one embodiment, the text matching unit 1505 is further used to:

[0201] Perform similarity matching processing on each commodity text feature information and each commodity feature information to obtain feature similarity between each commodity text feature information and each commodity feature information;

[0202] For each commodity text feature information, the feature similarity with the largest corresponding value is used as the feature matching value corresponding to each commodity text feature information.

[0203] In one embodiment, the text matching unit 1505 is further used to:

[0204] All feature matching values ​​are added together to obtain the matching result between the search text and the candidate product information;

[0205] Alternatively, weighted addition processing is performed on all feature matching values ​​to obtain a matching result between the text to be searched and the candidate product information.

[0206] In one embodiment, there are multiple candidate product information; the text matching unit 1505 is also used to:

[0207] Perform text matching processing on all product text feature information and product feature information corresponding to each candidate product information to obtain candidate matching results between the search text and each candidate product information;

[0208] According to the size of each candidate matching result, each candidate product information is sorted to obtain the matching result between the to-be-searched text and the candidate product information.

[0209] In one embodiment, the text matching unit 1505 is further used to:

[0210] When the matching result is that the text to be searched does not match the candidate product information, and the product feature information corresponding to the product key information is consistent with the product text feature information corresponding to the text key type, an error correction mark is added to the text to be searched and the candidate product information;

[0211] According to the error correction mark, the text to be searched and the candidate product information are subjected to secondary matching processing to obtain the secondary matching result of the text to be searched and the candidate product information;

[0212] The commodity feature information corresponds to commodity type information, the commodity type information includes commodity key information, and the commodity text type includes text key type.

[0213] In one embodiment, the text matching unit 1505 is further used to:

[0214] When the matching result is that the text to be searched does not match the candidate product information, and the product feature information corresponding to the product key information is consistent with the product text feature information corresponding to the text key type, an error correction mark is added to the text to be searched and the candidate product information;

[0215] According to the error correction mark, the text to be searched and the candidate product information are subjected to secondary matching processing to obtain the secondary matching result of the text to be searched and the candidate product information;

[0216] The commodity feature information corresponds to commodity type information, the commodity type information includes commodity key information, and the commodity text type includes text key type.

[0217] Reference Fig.16 The embodiment of the present invention further discloses a product matching device 1600, which includes:

[0218] at least one processor 1601;

[0219] At least one memory 1602, used to store at least one program;

[0220] When at least one program is executed by at least one processor 1601 , the commodity matching method of the previous embodiment is implemented.

[0221] An embodiment of the present invention further discloses a computer-readable storage medium, in which a computer program executable by a processor is stored. When the computer program executable by the processor is executed by the processor, it is used to implement the commodity matching method as described in the previous embodiment.

[0222] An embodiment of the present invention also discloses a computer program product, including a computer program or computer instructions, wherein the computer program or computer instructions are stored in a computer-readable storage medium, and a processor of a product matching device reads the computer program or computer instructions from the computer-readable storage medium, and the processor executes the computer program or computer instructions, so that the product matching device performs the product matching method as in the previous embodiment.

[0223] The terms "first", "second", "third", "fourth", etc. (if any) in the description of the present invention and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments of the present invention described herein can, for example, be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0224] It should be understood that in the present invention, "at least one (item)" refers to one or more, and "plurality" refers to two or more. "And / or" is used to describe the association relationship of associated objects, indicating that three relationships may exist. For example, "A and / or B" can represent: only A exists, only B exists, and A and B exist at the same time, where A and B can be singular or plural. The character " / " generally indicates that the previous and next associated objects are in an "or" relationship. "At least one of the following items" or similar expressions refers to any combination of these items, including any combination of single items or plural items. For example, at least one of a, b or c can represent: a, b, c, "a and b", "a and c", "b and c", or "a and b and c", where a, b, c can be single or multiple.

[0225] In the several embodiments provided by the present invention, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of units is only a logical function division. There may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be an indirect coupling or communication connection through some interfaces, devices or units, which can be electrical, mechanical or other forms.

[0226] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0227] In addition, each functional unit in each embodiment of the present invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0228] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product, which is stored in a storage medium and includes several instructions for a computer device (which can be a personal computer, a server or a network device, etc.) to perform all or part of the steps of the methods of each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (Read-Only Memory, referred to as ROM), random access memory (Random Access Memory, referred to as RAM), disk or optical disk and other media that can store program codes.

[0229] The step numbers in the above method embodiment are only provided for the convenience of explanation and description, and no limitation is imposed on the order of the steps. The execution order of each step in the embodiment can be adaptively adjusted according to the understanding of those skilled in the art.

Claims

1. A product matching method, It is characterized in that The following steps are involved: Acquire a text to be searched, perform text recall processing on the text to be searched, and obtain a plurality of product recall texts, each of which corresponds to a product text type; According to the commodity text type corresponding to each of the commodity recall texts, a commodity type label is added to each of the commodity recall texts to obtain a labeled commodity recall text; Performing text splicing on all the marked product recall texts to obtain the product texts to be processed, and performing feature extraction on the product texts to be processed to obtain product text feature information of each of the marked product recall texts; Acquire candidate product information preset in a candidate product link library according to the text to be searched, and acquire multiple product feature information of the candidate product information; Text matching processing is performed on all the commodity text feature information and all the commodity feature information to obtain a matching result between the text to be searched and the candidate commodity information.

2. The commodity matching method according to claim 1, It is characterized in that The text recall processing is performed on the text to be searched to obtain multiple product recall texts, including: Performing semantic feature recognition on the text to be searched to obtain multiple semantic feature information; Text recall processing is performed according to each of the semantic feature information to obtain multiple product recall texts.

3. The commodity matching method according to claim 2, It is characterized in that The text recall processing is performed according to each of the semantic feature information to obtain multiple product recall texts, including: Performing text recall processing according to each of the semantic feature information to obtain multiple candidate product recall texts; Obtaining multiple supplementary product recall texts according to the candidate product recall text and the candidate product information; A plurality of product recall texts are obtained according to all the candidate product recall texts and all the supplementary product recall texts.

4. The commodity matching method according to claim 3, It is characterized in that The step of obtaining a plurality of supplementary product recall texts according to the candidate product recall text and the candidate product information includes: Determining the type of supplementary product text corresponding to the candidate product recall text according to the candidate product recall text and the candidate product information; A plurality of supplementary product recall texts are obtained according to the supplementary product text type.

5. The commodity matching method according to claim 1, It is characterized in that The candidate product information includes a plurality of candidate product sub-information, and the candidate product sub-information corresponds to product type information; The commodity feature information is obtained by the following steps: According to the commodity type information, adding a candidate commodity type label to each of the candidate commodity sub-information to obtain labeled commodity information; All the marked commodity information are concatenated to obtain the commodity information to be processed, and feature extraction is performed on the commodity information to be processed to obtain commodity feature information of each of the marked commodity information.

6. The commodity matching method according to claim 1, It is characterized in that The performing text matching processing on all the commodity text feature information and all the commodity feature information to obtain a matching result between the to-be-searched text and the candidate commodity information includes: Performing text matching processing on each of the commodity text feature information and all of the commodity feature information to obtain a feature matching value corresponding to each of the commodity text feature information; According to all the feature matching values, a matching result between the text to be searched and the candidate product information is obtained.

7. The commodity matching method according to claim 6, It is characterized in that The step of performing text matching processing on each of the commodity text feature information and all of the commodity feature information to obtain a feature matching value corresponding to each of the commodity text feature information includes: Performing similarity matching processing on each of the commodity text feature information and each of the commodity feature information to obtain feature similarities between each of the commodity text feature information and each of the commodity feature information; For each of the commodity text feature information, the feature similarity with the largest corresponding value is used as the feature matching value corresponding to each of the commodity text feature information.

8. The commodity matching method according to claim 6, It is characterized in that The obtaining, according to all the feature matching values, a matching result between the text to be searched and the candidate product information includes: Adding all the feature matching values ​​to obtain a matching result between the text to be searched and the candidate product information; Alternatively, weighted addition processing is performed on all the feature matching values ​​to obtain a matching result between the text to be searched and the candidate product information.

9. The commodity matching method according to claim 1, It is characterized in that There are multiple candidate product information; performing text matching processing on all the product text feature information and all the product feature information to obtain a matching result between the text to be searched and the candidate product information includes: Performing text matching processing on all the product text feature information and the product feature information corresponding to each of the candidate product information to obtain candidate matching results between the text to be searched and each of the candidate product information; According to the size of each candidate matching result, each candidate product information is sorted to obtain the matching result between the text to be searched and the candidate product information.

10. The commodity matching method according to claim 1, It is characterized in that The commodity feature information corresponds to commodity type information, the commodity type information includes commodity key information, and the commodity text type includes text key type; After obtaining the matching result between the text to be searched and the candidate product information, the product matching method further includes: When the matching result is that the text to be searched does not match the candidate product information, and the product feature information corresponding to the product key information is consistent with the product text feature information corresponding to the text key type, adding an error correction mark to the text to be searched and the candidate product information; According to the error correction mark, the text to be searched and the candidate product information are subjected to secondary matching processing to obtain a secondary matching result between the text to be searched and the candidate product information.

11. The commodity matching method according to claim 1, It is characterized in that The commodity feature information corresponds to commodity type information, the commodity type information includes commodity key information, and the commodity text type includes text key type; After obtaining the matching result between the text to be searched and the candidate product information, the product matching method further includes: When the matching result is that the text to be searched does not match the candidate product information, and the product feature information corresponding to the product key information is consistent with the product text feature information corresponding to the text key type, adding an error correction mark to the text to be searched and the candidate product information; According to the error correction mark, the text to be searched and the candidate product information are subjected to secondary matching processing to obtain a secondary matching result between the text to be searched and the candidate product information.

12. A commodity matching device, It is characterized in that include: A text recall unit, used to obtain a text to be searched, perform text recall processing on the text to be searched, and obtain a plurality of product recall texts, each of which corresponds to a product text type; a label adding unit, configured to add a product type label to each of the product recall texts according to the product text type corresponding to each of the product recall texts, so as to obtain a marked product recall text; A feature extraction unit is used to perform text splicing on all the marked product recall texts to obtain a product text to be processed, and perform feature extraction on the product text to be processed to obtain product text feature information of each of the marked product recall texts; An information acquisition unit, configured to acquire candidate commodity information preset in a candidate commodity link library according to the text to be searched, and acquire a plurality of commodity feature information of the candidate commodity information; The text matching unit is used to perform text matching processing on all the commodity text feature information and all the commodity feature information to obtain a matching result between the text to be searched and the candidate commodity information.

13. A commodity matching device, It is characterized in that include: at least one processor; at least one memory for storing at least one program; When at least one of the programs is executed by at least one of the processors, the commodity matching method according to any one of claims 1 to 11 is implemented.

14. A computer-readable storage medium, It is characterized in that A computer program executable by a processor is stored therein, and when the computer program executable by the processor is executed by the processor, it is used to implement the commodity matching method according to any one of claims 1 to 11.

15. A computer program product comprising a computer program or computer instructions, It is characterized in that The computer program or the computer instructions are stored in a computer-readable storage medium, and the processor of the product matching device reads the computer program or the computer instructions from the computer-readable storage medium. The processor executes the computer program or the computer instructions, so that the product matching device performs the product matching method as described in any one of claims 1 to 11.

Images