Information processing method and device, equipment and storage medium

Abstract

The application discloses an information processing method, device and equipment and a storage medium. The information processing method comprises the following steps: obtaining merchant information of a merchant, wherein the merchant information comprises a merchant name and a merchant operation category; then, obtaining a brand name matched with the merchant name according to the merchant name; further, generating a candidate brand set corresponding to the merchant according to the brand name in the case that the merchant operation category and the brand name are matched with a corresponding brand operation category; and then, determining target brand information corresponding to the merchant in the candidate brand set through a merchant brand association determination model. In this way, the merchant short name information and the merchant category information are introduced into the matching process of the merchant and the brand, so that the probability of merchant missing association and wrong association caused by full-word matching of the merchant name and the brand name can be effectively reduced, and the accuracy of association between the merchant and the brand can be improved.

Description

Technical Field

[0001] This application belongs to the field of information processing technology, and in particular relates to an information processing method, apparatus, device and storage medium. Background Technology

[0002] Merchants refer to enterprises, shops, associations, and other legally established organizations that have physical business premises or operate on online platforms. To facilitate the management of merchant information by various service platforms, merchant names are sometimes associated with their corresponding brand names to obtain the relationship between the merchant and the brand.

[0003] In related technologies, some brand names are not included in merchant names, and some brand names are short or common and may appear in many unrelated merchant names. Therefore, full-word matching based on merchant names and brand names may result in false or false associations, affecting the accuracy of the association between merchants and their brands. Summary of the Invention

[0004] This application provides an information processing method, apparatus, device, and storage medium that can solve the problem of low accuracy in associating merchants with their brands in the prior art.

[0005] In a first aspect, embodiments of this application provide an information processing method, which may include:

[0006] Obtain merchant information, including merchant name and business category;

[0007] Based on the merchant name, obtain the brand name that matches the merchant name, and the brand name corresponds to the brand's business category;

[0008] When the merchant's business category matches the brand's business category, a set of candidate brands corresponding to the merchant is generated based on the brand name. The set of candidate brands includes candidate brand information for at least one candidate brand.

[0009] The merchant brand association determination model identifies the target brand information corresponding to the merchant from the candidate brand set. The target brand information includes the brand name corresponding to the merchant name.

[0010] Secondly, embodiments of this application provide an information processing apparatus, which may include:

[0011] The acquisition module is used to acquire merchant information, which includes the merchant name and the merchant's business category.

[0012] The processing module is used to obtain the brand name that matches the merchant name, and the brand name corresponds to the brand's business category.

[0013] The generation module is used to generate a set of candidate brands corresponding to a merchant based on the brand name when the merchant's business category matches the brand's business category. The set of candidate brands includes candidate brand information for at least one candidate brand.

[0014] The determination module is used to determine the target brand information corresponding to the merchant from the candidate brand set through the merchant brand association determination model. The target brand information includes the brand name corresponding to the merchant name.

[0015] Thirdly, embodiments of this application provide a computing device, which includes: a processor and a memory storing computer program instructions;

[0016] When the processor executes computer program instructions, it implements the information processing method as described in the first aspect.

[0017] Fourthly, embodiments of this application provide a computer storage medium storing computer program instructions, which, when executed by a processor, implement the information processing method as described in the first aspect.

[0018] Fifthly, embodiments of this application provide a chip, which includes a processor and a communication interface. The communication interface and the processor are coupled, and the processor is used to run programs or instructions to implement the information processing method as shown in the first aspect.

[0019] In a sixth aspect, embodiments of this application provide a computer program product stored in a storage medium, which is executed by at least one processor to implement the information processing method as described in the first aspect.

[0020] The information processing method, apparatus, device, and storage medium of this application embodiment acquire merchant information, including merchant name and merchant business category. Then, based on the merchant name, a brand name matching the merchant name is acquired. If the merchant business category and the brand name match the corresponding brand business category, a candidate brand set corresponding to the merchant is generated based on the brand name. The candidate brand set includes candidate brand information of at least one candidate brand. Then, through a merchant brand association determination model, the target brand information corresponding to the merchant is determined from the candidate brand set. In this way, by introducing merchant abbreviation information and merchant category information into the matching process between merchants and brands, the probability of missed or incorrect merchant association caused by full-word matching of merchant name and brand name can be effectively reduced, the accuracy of merchant-brand association can be improved, and thus effective management of merchants can be achieved. In addition, the embodiments of this application do not require targeted crawling of keywords in each brand platform webpage, nor do they require obtaining the keywords and non-keywords of each brand. It is only necessary to construct a dataset containing merchant and brand names, as well as the business categories of merchants and brands, to complete the association between merchants and their brands. In this way, the requirements for brand information are lower and easier to obtain, reducing the cost and complexity of the process of associating merchants with their brands. Attached Figure Description

[0021] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the embodiments of this application will be briefly introduced below. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0022] Figure 1 This is a schematic diagram of an information processing architecture according to an embodiment of the information processing method provided in this application;

[0023] Figure 2 This is a second schematic diagram of an information processing architecture according to an embodiment of the information processing method provided in this application;

[0024] Figure 3 A flowchart illustrating an information processing method provided in an embodiment of this application;

[0025] Figure 4 A flowchart illustrating the first sample set of an information processing method provided in an embodiment of this application;

[0026] Figure 5 A flowchart illustrating the second sample set of an information processing method provided in an embodiment of this application;

[0027] Figure 6 A schematic diagram of a merchant brand association determination model for an information processing method provided in this application embodiment;

[0028] Figure 7 This is a schematic diagram of the structure of an information processing apparatus provided in one embodiment of this application;

[0029] Figure 8 This is a schematic diagram of the structure of an information processing device provided in one embodiment of this application. Detailed Implementation

[0030] The features and exemplary embodiments of various aspects of this application will be described in detail below. To make the objectives, technical solutions, and advantages of this application clearer, the application will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are only intended to explain this application and not to limit it. For those skilled in the art, this application can be implemented without some of these specific details. The following description of the embodiments is merely to provide a better understanding of this application by illustrating examples.

[0031] It should be noted that, in this document, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising..." does not exclude the presence of additional identical elements in the process, method, article, or apparatus that includes the element.

[0032] Among all merchant-related information, brand information corresponding to the merchant's brand is always a crucial aspect of merchant management. Compared to individual merchants, branded chain merchants are relatively stable and typically widely distributed, facilitating management by various service platforms. In some cases, when providing new services or adjusting business operations, service platforms need to analyze massive amounts of merchant information. However, the merchant information obtained by these platforms is often incomplete, and they lack channels to supplement it. Therefore, acquiring the brand information corresponding to these merchants can supplement the missing merchant information from multiple angles and perspectives. Thus, accurately associating merchants with their brands to obtain and manage merchant information has become a pressing issue for various service platforms.

[0033] In one example of related technologies, a service platform can perform full-word matching of merchant and brand names based on a merchant and brand dictionary. However, some brand names are short or generic, and these brand names may appear in many unrelated merchant names, leading to false associations. For example, the home furnishing brand IKEA is a generic name, and therefore appears in merchant names such as "IKEA Floor Heating Equipment Store in City A" and "IKEA Home Furnishings in County B". Thus, associating merchants and brands in this way can lead to false associations between such merchants and unrelated brands, reducing the accuracy of merchant-brand associations. Furthermore, since some merchants submit their names using abbreviations, using the full name for full-word matching can also lead to association failures. For example, the service platform contains numerous merchant names such as "Yonghui (Jixian Store)" and "Yonghui (Jinshan District Jinshan Wanda Plaza Store)," which omit the "supermarket" field. Therefore, the full-word matching method will miss such merchants. Additionally, some merchants on various service platforms typically submit their registered names, resulting in some brand names not being fully included in the merchant names. For instance, the merchant names "Shuyu Pingmin Pharmacy" and "Heze City Mudan District Shuyu Health Service Pharmacy" are not entirely included in the brand name, as are the merchant names "Ruian Rong'an Real Estate Co., Ltd." and the brand name "Rong'an Real Estate." Thus, the full-word matching method will also miss such merchants, reducing the accuracy of merchant-brand association.

[0034] In another example, the service platform can also associate merchants with their brands by constructing a probability matrix based on keywords in the brand name. Specifically, a brand dictionary is built using web crawlers, establishing a keyword database of regions, brands, and industries related to and unrelated to the brand. Then, a probability matrix is ​​constructed based on the text matching results to calculate the part-of-speech of each keyword in the merchant's name. Finally, the association between the merchant and the brand is achieved by matching keywords. However, in the process of building the keyword database, it is necessary to crawl keywords from the brand platform's web pages. Different crawlers need to be built for different brand platform web pages, which not only increases the cost of obtaining information but also makes the process of building the keyword database cumbersome and time-consuming. In addition, sometimes the content of some brand platform web pages is missing, making it impossible to obtain the keywords from the brand platform web pages comprehensively and accurately. Furthermore, some keywords may have synonyms. Therefore, it is difficult to obtain accurate merchant and brand information using the above method, which will also reduce the accuracy of the association between the merchant and the brand, leading to matching failures.

[0035] Based on this, in order to solve the aforementioned problems, this application provides an information processing method that incorporates merchant and brand names, as well as merchant and brand operating categories, into the merchant-brand matching process. Specifically, it obtains merchant information, including the merchant name and business category. Then, based on the merchant name, it obtains the brand name that matches the merchant name. If the business category and brand name match the corresponding brand operating category, it generates a candidate brand set corresponding to the merchant based on the brand name. This candidate brand set includes candidate brand information for at least one candidate brand. Then, based on a merchant-brand association determination model trained using a weakly supervised sample set and a manually labeled sample set, it determines the target brand information corresponding to the merchant. Thus, by incorporating merchant abbreviation information and merchant category information into the merchant-brand matching process, it can effectively reduce the probability of missed or incorrect merchant associations caused by full-word matching of merchant and brand names, improve the accuracy of merchant-brand association, and thereby achieve effective merchant management. In addition, the embodiments of this application do not require targeted crawling of keywords in each brand platform webpage, nor do they require obtaining the keywords and non-keywords of each brand. It is only necessary to construct a dataset containing merchant and brand names, as well as the business categories of merchants and brands, to complete the association between merchants and their brands. In this way, the requirements for brand information are lower and easier to obtain, reducing the cost and complexity of the process of associating merchants with their brands.

[0036] Based on this, embodiments of this application provide an information processing method, apparatus, device, and storage medium. The following will describe in conjunction with the appendix... Figures 1 to 6 This application describes in detail the information processing methods, apparatus, servers, and storage media of the embodiments thereof. It should be noted that these embodiments are not intended to limit the scope of this application.

[0037] First, one of the information processing architectures of the information processing method provided in the embodiments of this application will be described.

[0038] like Figure 1 As shown, the information processing architecture 10 may include a merchant information and brand information construction module 101, a merchant information and brand information preprocessing module 102, a merchant brand weakly supervised dataset construction module 103, a merchant pre-trained model task module 104, a merchant brand pre-trained model training module 105, a merchant brand association determination module 106, and a merchant brand association prediction module 107. The following is a detailed description of each module in its information processing architecture.

[0039] The merchant information and brand information construction module 101 is used to collect information on real-world brands and merchants, ensuring that this information is broad and general. Specifically, it collects information on multiple brands to construct a brand information set and acquires merchant information to construct the aforementioned information set. It also transmits information to the merchant information and brand information preprocessing module 102, providing it with preprocessed merchant and brand information.

[0040] Here, the brand information set can include brand information from multiple brands. Specifically, the brand information for each brand can include at least one of the following: brand name, brand business category, and brand identity identifier. For example, a brand dataset could be: {1: SF Express, 2: Vienna Hotel, 3: Xiaolongkan Hot Pot, 4: Yonghui Supermarket, 5: Shuyu Pharmacy, 6: IKEA}, where the numbers represent the brand identity identifier. Furthermore, the brand name includes at least one of the following: full brand name, brand abbreviation, and brand identity identifier.

[0041] It should be noted that the brand in this application embodiment can refer to a socially recognized identifier that embodies standardized and high-quality products. Specifically, the brand can be the name of a merchant or platform, a product trademark, or a service name. The full name (or abbreviation) of the brand in this application embodiment can be the full name (or abbreviation) in Chinese, or the full name (or abbreviation) in other languages, such as the full name (or abbreviation) in English. The brand business category in this application embodiment refers to the methods and approaches adopted by the merchant or service platform in its business activities, such as mining, manufacturing, wholesale, retail, consulting, leasing, agency, etc. The brand identity identifier in this application embodiment can be a mark used by the merchant information and brand information construction module 101 to distinguish a large number of brands.

[0042] A merchant information set may include merchant information for multiple merchants. Merchant information may include at least one of the following: merchant name, merchant business category, and merchant identification. For example, a merchant dataset might include: [Vienna International Hotel (Xiabu Road Branch), Heze City Mudan District Shuyu Health Service Pharmacy, Yonghui (Jixian Branch), Zhuji City IKEA Floor Heating Equipment Store].

[0043] It should be noted that the full name (or abbreviation) of a merchant can be in Chinese (or Chinese abbreviation) or in other languages ​​(or English abbreviations). In this embodiment, the merchant's business category refers to the methods and approaches adopted by the merchant or service platform in its business activities, such as mining, manufacturing, wholesale, retail, consulting, leasing, and agency in industries like manufacturing, handicrafts, construction, transportation, commerce, catering, services, and repair. In this embodiment, the merchant identity identifier can be a marker used by the merchant information and brand information construction module 101 to distinguish a large number of merchants.

[0044] The merchant information and brand information preprocessing module 102 is used to preprocess each piece of merchant information and brand information, using preprocessing algorithms such as merchant abbreviation extraction algorithm, brand abbreviation extraction algorithm, merchant business category identification algorithm, and expert knowledge to obtain preprocessed information for merchant and brand association; and it is used to transmit information with the merchant brand weakly supervised dataset construction module 103 and the merchant brand association prediction module 107, providing them with preprocessed merchant information and brand information to construct a merchant brand weakly supervised dataset.

[0045] Specifically, in one example, the merchant information and brand information preprocessing module 102 can also be used to extract all abbreviations for each merchant's full name using the aforementioned merchant abbreviation extraction algorithm, for example, merchant abbreviations: [Vienna, Shuyu, Yonghui, IKEA]; or, using the aforementioned brand abbreviation extraction algorithm, to extract all abbreviations for each brand's full name, for example, brand abbreviations: [SF Express, Vienna, Xiaolongkan, Yonghui, Shuyu, IKEA]. Here, for ease of description, it could be to extract all Chinese and English abbreviations for each Chinese brand's full name.

[0046] In another example, the merchant information and brand information preprocessing module 102 can also be used to establish a brand name mapping dictionary. For each brand, both the full brand name (including full names in Chinese, English, and other languages) and the brand abbreviation (including abbreviations in Chinese, English, and other languages) are added to the brand name mapping dictionary, and these brand names are mapped to the brand identity identifier of that brand. For example, the brand dictionary could be: {SF Express:1, Vienna:2, Xiaolongkan:3, Yonghui:4, Shuyu:5, IKEA:6, IKEA:6}, where the numbers represent the brand identity identifier.

[0047] In another example, the merchant information and brand information preprocessing module 102 can also be used to obtain the merchant business category for each merchant when the merchant information does not include the merchant business category, using the aforementioned merchant business category identification algorithm. For example, merchant business categories: {Vienna International Hotel (Xiabu Road Branch): Hotel, Heze Mudan District Shuyu Health Service Pharmacy: Pharmacy, Yonghui (Jixian Branch): Supermarket, Zhuji IKEA Floor Heating Equipment Store: Home Decoration}. Based on this, the brand business category of the brand information is similar, for example: {SF Express: Life Services, Vienna Hotel: Hotel, Xiaolongkan Hot Pot: Chinese Food, Yonghui Supermarket: Supermarket, Shuyu Pharmacy: Pharmacy, IKEA: Home Decoration}.

[0048] In another example, the merchant information and brand information preprocessing module 102 can also be used to map the merchant business category and the brand business category using the expert knowledge mentioned above, so as to obtain the mapping relationship between the merchant business category and the brand business category, so as to ensure that the merchant business category can be mapped to the brand business category.

[0049] In another example, the merchant information and brand information preprocessing module 102 can also be used to map the merchant business category to the brand business category for each merchant based on the merchant business category and brand business category mapping relationship established above, so as to apply to the subsequent process of associating merchant information with brand information.

[0050] The merchant brand weakly supervised dataset construction module 103 is used to make preliminary associations between merchant brands using merchant and brand abbreviations and categories, obtain a candidate brand set, and construct a merchant brand weakly supervised dataset based on the candidate brand set; and to interact with the merchant pre-trained model task module 104, the merchant brand association determination module 106, and the merchant brand association prediction module 107.

[0051] Specifically, the merchant brand weakly supervised dataset construction module 103 is used to establish an initial candidate brand set for each merchant, which is initially empty. Next, for each merchant's abbreviation, a search is performed in the brand name mapping dictionary. If a matching brand abbreviation is found, it is further determined whether the merchant's business category and the brand's corresponding business category are the same. If they are the same, the brand information, such as the brand name (including the brand abbreviation and full brand name) and its corresponding brand identity pair, is added to the merchant's initial candidate brand set, thus obtaining the merchant's candidate brand set. Then, repeat the above process until each merchant has been processed. In this way, a merchant brand dataset can be constructed based on the candidate brand set of each merchant, that is, a merchant brand weakly supervised dataset. For example, [Vienna International Hotel (Xiabu Road Store): [(Vienna Hotel, 2)], Heze City Mudan District Shuyu Health Service Pharmacy: [(Shuyu Pharmacy, 5)], Yonghui (Jixian Store): [(Yonghui Supermarket, 4)], Zhuji City IKEA Floor Heating Equipment Store: [(IKEA), 6)]], where the numbers represent the brand identity identifiers of the brands.

[0052] The merchant pre-training model task module 104 is used to define two types of merchant pre-training tasks, which are used to build a sample set for the subsequent merchant brand pre-training model training module 105 to train the model.

[0053] In one example, the merchant pre-training model task module 104 can specifically be used to mask key information in the full name of a merchant, enabling the first preset model to learn the semantic relationships between the merchant name, the sample merchant's business category, and the sample brand information. Since some content in the merchant information, such as the merchant's abbreviation and detailed address information, has a weak relationship with other contextual content, masking and predicting this content has low value. Therefore, unlike general masked language model (MLM) pre-training tasks that mask all text in the text with a certain probability, this embodiment of the application adopts a merchant text information masking strategy based on weakly supervised data. This strategy makes the first preset model focus on the suffix text after the merchant's abbreviation in the merchant name, allowing the first preset model to learn merchant information that is more relevant to the merchant's brand and category. For example, if the input sample sequence of the first preset model is [CLS]Heze City Mudan District Shuyu Health Service[MASK][MASK][MASK][SEP]No. 5, Zhongda Yijing Plaza, No. 1007 Chongqing Road, Mudan District, Heze City, Shandong Province[SEP]Pharmacy[SEP]Shuyu Pharmacy, the purpose of training is to make the first preset model output the three characters "Pharmacy" at the [MASK] symbol position.

[0054] In another example, the merchant pre-trained model task module 104 can be used to construct positive and negative examples using the candidate brand set for each merchant in the weakly supervised merchant brand dataset, enabling the second pre-defined model to learn the semantic association between merchant brands and merchant names. Unlike the current method of directly and randomly replacing brands to construct negative examples, this embodiment introduces merchant abbreviations, which distinguishes between the abbreviation and the suffix of the full merchant name, and randomly replaces each part with a certain probability. This method can more effectively generate negative example samples with higher relevance, thereby improving the accuracy of merchant brand matching. For example, the input sample sequence of the second preset model is [CLS]Vienna International Hotel (Xiabu Road Branch) [SEP]No.39, Xiabu Road, Huicheng District, Huizhou City, Guangdong Province [SEP]Hotel [SEP]SF Express Hotel and [CLS]Vienna International Hotel (Xiabu Road Branch) [SEP]No.39, Xiabu Road, Huicheng District, Huizhou City, Guangdong Province [SEP]Hotel [SEP]Vienna Hot Pot. The purpose of training is to make the second preset model output 0 at the [CLS] symbol position, which indicates that the two are not related.

[0055] The merchant brand pre-training model training module 105 is used to train the first preset model and the second preset model based on the sample set output by the merchant pre-training model task module 104, to obtain an initial first decision sub-model corresponding to the first preset model and an initial second decision sub-model corresponding to the second preset model. Based on the initial first and second decision sub-models, a merchant brand pre-training model is generated. The first and second preset models can be Chinese pre-trained language models such as dynamically fused entity recognition models like RoBERTa-wwm, RoBERTa, and ERNIE2; they are used for data interaction with the merchant brand association determination module 106 and the merchant brand association prediction module 107. Then, the initial first and second decision sub-models can be trained alternately in batches, using cross-entropy as the error and the Adam optimizer for model optimization. Each training iteration trains on the full dataset of both task training sets, and training stops after reaching the preset maximum number of iterations, resulting in the merchant brand pre-training model.

[0056] It should be noted that the Chinese pre-trained language models mentioned above can be replaced with any well-trained pre-trained language model that supports character-level Chinese text, such as Chinese-BERT, Chinese-Electra, Chinese-ALBERT, etc., without limitation. The Adam optimizer can also be replaced with optimization methods such as L-BFGS, SGD, RMSProp, etc., without limitation.

[0057] The merchant brand association determination module 106 is used to fine-tune the downstream task based on the above-mentioned merchant brand pre-trained model using a small-scale manually labeled sample set, thereby obtaining the merchant brand association determination model; it is used to interact with the merchant brand pre-trained model training module 105 and the merchant brand association prediction module 107.

[0058] In one example, to obtain a small-scale manually labeled sample set, this embodiment of the application can use the merchant brand weakly supervised dataset construction module 103 described above to associate a batch of new merchant information and brand information to obtain new candidate brand information. Then, the matching of merchant and brand pairs in each candidate brand information is determined by manual verification. During the labeling process, three labeling staff members label the merchant and brand pairs respectively, and select the majority labeling results as the final manual labeling result. The final manual labeling result can include manual association information, which includes the relationship pairs between manual merchant information and manual brand information. For example, the sample sequence input to the merchant brand pre-training model is [CLS] Zhuji City IKEA Floor Heating Equipment Business Department [SEP] No. 72-15, Gaohu Road, Jiyang Street, Zhuji City, Shaoxing City, Zhejiang Province [SEP] Home Decoration [SEP] IKEA. The purpose of training is to output 0 at the [CLS] symbol position, which indicates that the two are not related.

[0059] The merchant brand association prediction module 107 is used to associate merchants and brands based on the output model of at least one of the following modules: the merchant brand pre-training model training module 105 and the merchant brand association determination module 106.

[0060] Specifically, for each input merchant, the merchant information and brand information preprocessing module 102 is used to obtain the merchant information, which includes the merchant name and the merchant business category. The brand name matching the merchant name is obtained based on the merchant abbreviation in the merchant name. The brand name corresponds to the brand business category. Next, the merchant brand weakly supervised dataset construction module 103 is used to construct a candidate brand set corresponding to the merchant. Then, the merchant and brand are associated based on the output model of at least one of the following modules: the merchant brand pre-training model training module 105 and the merchant brand association determination module 106.

[0061] Based on the above architecture, the following will be combined with, for example Figure 2 The information processing architecture shown herein will be used to provide a detailed description of the information processing methods provided in the embodiments of this application.

[0062] like Figure 2 As shown, the information processing architecture 10 can receive merchant A sent by the user terminal and send the merchant name A to the merchant information and brand information preprocessing module 102.

[0063] The merchant and brand information preprocessing module 102 retrieves merchant information corresponding to merchant A, including the merchant name and business category. The merchant name can include both the full name and abbreviation. Next, based on the merchant name (which can be the abbreviation), the module retrieves the matching brand name (which can be the abbreviation), where the brand name corresponds to the brand's business category. Then, the module 102 sends the merchant and brand information to the merchant brand weakly supervised dataset construction module 103.

[0064] The merchant brand weakly supervised dataset construction module 103 matches the merchant business category in the merchant information with the brand business category in the brand information. When the merchant business category and brand business category match, it generates a candidate brand set corresponding to the merchant based on the brand name. The candidate brand set includes candidate brand information of at least one candidate brand. Then, the merchant brand weakly supervised dataset construction module 103 combines its candidate brands and sends them to the merchant brand association prediction module 107. For example, for merchant A: Yonghui (Jixian Store), it first obtains its merchant abbreviation "Yonghui" and merchant business category "supermarket", and matches it with the brand name "Yonghui Supermarket". Therefore, it constructs the information for inputting the merchant brand association judgment model: [CLS] Yonghui (Jixian Store) [SEP] New Century Jixian Shopping Plaza, Intersection of Jingjin Highway and Yanji Road, Beicang Town, Beichen District, Tianjin [SEP] Supermarket [SEP] Yonghui Supermarket, where [CLS] and [SEP] are used as placeholders for brand association judgment and to separate different parts of the input, respectively.

[0065] The merchant brand association prediction module 107 uses a merchant brand association determination model to determine the target brand information corresponding to the merchant from the candidate brand set. The target brand information includes the brand name corresponding to the merchant name. Then, the merchant brand association prediction module 107 outputs the association result. For example, outputting 1 at the [CLS] symbol position indicates that the merchant "Yonghui (Jixian Store)" is associated with the brand "Yonghui Supermarket".

[0066] Therefore, the information processing method provided in this application, by introducing merchant abbreviation information and merchant category information into the merchant-brand matching process, can effectively reduce the probability of missed or incorrect merchant associations caused by full-word matching of merchant names and brand names, improve the accuracy of merchant-brand association, and thus achieve effective merchant management. Furthermore, this application does not require targeted crawling of keywords from each brand platform webpage, nor does it require obtaining keywords and non-keywords for each brand. It only needs to construct a dataset containing merchant and brand names, as well as merchant and brand business categories, to complete the association between merchants and brands. This reduces the requirements for brand information, makes it easier to obtain, and lowers the cost and complexity of the merchant-brand association process.

[0067] It should be noted that the information processing method provided in this application embodiment can be applied not only to the above-mentioned scenario of matching brand name based on merchant name, but also to the scenario of matching merchant name based on brand name. The information processing method in this scenario is the same in principle as the information processing method in the above-mentioned scenario of matching brand name based on merchant name, and will not be described in detail here. This application embodiment is used as an example to illustrate the scenario of matching brand name based on merchant name, and does not limit its application scenario.

[0068] Based on the above information processing architecture and application scenarios, the following will combine... Figure 3 The information processing method provided in the embodiments of this application will be described in detail.

[0069] Figure 3 This is a flowchart of an information processing method provided in an embodiment of this application.

[0070] like Figure 3 As shown, this information processing method can be applied to, for example... Figure 1 The information processing architecture shown may include the following steps in its specific information processing method:

[0071] Step 310: Obtain the merchant's information, including the merchant's name and business category; Step 320: Obtain the brand name matching the merchant's name, where the brand name corresponds to the brand's business category; Step 330: If the merchant's business category matches the brand's business category, generate a candidate brand set corresponding to the merchant based on the brand name, where the candidate brand set includes candidate brand information for at least one candidate brand; Step 340: Using a merchant brand association determination model, determine the target brand information corresponding to the merchant from the candidate brand set, where the target brand information includes the brand name corresponding to the merchant's name.

[0072] Therefore, introducing merchant abbreviation and category information into the merchant-brand matching process can effectively reduce the probability of missed or incorrect merchant associations caused by full-word matching of merchant and brand names, improve the accuracy of merchant-brand association, and thus achieve effective merchant management. Furthermore, this embodiment does not require targeted crawling of keywords from each brand platform webpage, nor does it require obtaining keywords and non-keywords for each brand. It only needs to construct a dataset containing merchant and brand names, as well as merchant and brand business categories, to complete the association between merchants and brands. This reduces the requirements for brand information, makes it easier to obtain, and lowers the cost and complexity of the merchant-brand association process.

[0073] The above steps are explained in detail below:

[0074] First, step 310 involves obtaining the merchant's information, which includes the merchant's name and business category. At this point, the merchant's name may include the full name and abbreviation of the merchant.

[0075] Next, regarding step 320, in one or more possible embodiments, if the merchant name is a merchant abbreviation, step 320 may specifically include:

[0076] The merchant's abbreviation is matched with the brand abbreviations in the preset brand information set to obtain the matching results;

[0077] If the matching results indicate that there is a brand abbreviation corresponding to the merchant abbreviation in the preset brand information set, the brand information of the brand abbreviation corresponding to the merchant abbreviation is obtained. The brand information includes the brand name and brand business category.

[0078] Furthermore, regarding step 330, in one or more possible embodiments, the brand information also includes the brand's brand identity identifier. Based on this, step 330 may specifically include:

[0079] If the merchant's business category matches the brand's business category, obtain the initial set of candidate brands corresponding to the merchant;

[0080] Add the brand name and brand identity to the initial candidate brand set to obtain the candidate brand set corresponding to the merchant.

[0081] Then, regarding step 340, in one or more possible embodiments, prior to step 340, the information processing method may further include a process of training a merchant brand association determination model, as detailed below.

[0082] Step 3501: Obtain a sample set, which includes a first sample set and a second sample set. The first sample set includes information used to characterize the semantic association between the sample merchant name, the sample merchant business category and the sample brand information. The second sample set includes information used to characterize the semantic association between the sample merchant name and the sample brand information.

[0083] Step 3502: Train the first preset model based on the first sample set, and obtain an initial first decision sub-model when the first preset model meets the first preset training conditions; and train the second preset model based on the second sample set, and obtain an initial second decision sub-model when the second preset model meets the second preset training conditions; here, the second preset training conditions include at least one of the following: the number of iterations of the second decision sub-model is greater than or equal to the second preset number of iterations, and the probability of the model outputting merchant-brand association at a preset position is greater than the first preset probability value; Step 3503: Generate a merchant-brand pre-trained model based on the initial first decision sub-model and the initial second decision sub-model;

[0084] Step 3504: Obtain the manually labeled sample set. The sample set includes multiple manually associated information, which includes relationship pairs between personal business information and manually labeled brand information.

[0085] Step 3505: Train the merchant brand pre-training model based on multiple artificial association information. If the merchant brand training model meets the third preset training condition, obtain the merchant brand association judgment model. The third preset training condition includes at least one of the following: the number of iterations of the merchant brand training model is greater than or equal to the third preset iteration number, and the probability of the model outputting merchant-brand association at a preset position is greater than the second preset probability value.

[0086] The following is a detailed explanation of steps 3501 to 3505 above, as shown below:

[0087] First, regarding step 3501, in one instance, the sample merchant name may include the full name of the sample merchant. Based on this, the first sample set involved in step 3501 can be obtained through the following steps 35011 to 35014.

[0088] Step 35011: Perform word segmentation on the full name of the sample merchant to obtain a full name word sequence. The full name word sequence includes N words, where N is a positive integer.

[0089] Step 35012: Mark each word in the N words using a preset marking algorithm to obtain a marking sequence. The marking sequence includes a first identifier and a second identifier. The first identifier is used to identify the first word in the N words that can be masked, and the second identifier is used to identify the second word in the N words that cannot be masked.

[0090] Specifically, the sample merchant name also includes j sample merchant abbreviations, which are determined by the text in the full name of the sample merchant; the first word is the word after the first abbreviation among the j sample merchant abbreviations that does not belong to the j sample merchant abbreviations, where j is a positive integer.

[0091] Step 35013: The first word is masked using a preset masking algorithm to obtain the masked name of the sample merchant.

[0092] Specifically, when the first word includes i characters, where i is a positive integer greater than 1, the first preset number of characters in the i characters are replaced with preset characters, and the second preset number of characters in the i characters are replaced with random characters; according to the order of the i characters, the third preset number of original characters, preset characters and random characters in the i characters are concatenated to obtain the sample merchant mask name.

[0093] Step 35014: The sample merchant mask name, sample merchant name, sample merchant identity identifier, sample merchant business category and sample brand name are determined as the first sample set, and the sample merchant identity identifier is the identity identifier of the sample merchant corresponding to the sample merchant name.

[0094] It should be noted that the first preset training conditions include at least one of the following: the number of iterations of the first decision sub-model is greater than or equal to the first preset number of iterations, and the decision result output by the first decision sub-model includes the first word of the sample merchant mask name before it is masked.

[0095] For example, such as Figure 4 As shown, for each merchant's full name, such as "Heze City Mudan District Shuyu Health Service Pharmacy", a word segmentation tool is used to segment the merchant's full name into words, thus obtaining a sample full name word sequence W = [w1, w2, ..., w aj ,…w N ], where w1 represents Heze City, w2 represents Mudan District, w3 represents Shuyu, w4 represents health, w5 represents service, and w6 represents a pharmacy. N represents the sequence length, which is also the number of words; here, it is also a j This represents the index of the j-th merchant abbreviation. For each word w in this word sequence... i Assign it the label y i This is used to identify whether the word can be masked. For each word w i If this word is located after the first merchant abbreviation (i.e., i > a1) and is not a word in the merchant abbreviation, then y i =1 indicates that this word can be masked; otherwise, y i=0, it indicates that this word cannot be masked, thereby obtaining the tag sequence Y=[y1,y2,…y N , for example Y=[0,0,0,1,1,1]. Then, in the masking step, a word with y i =1 and not masked yet is randomly selected for masking each time. The masking strategy is that for the word w i each character, there is an 80% probability of replacing it with the preset character "[MASK]", for example, "药" is replaced with "[MASK]"; a 10% probability of replacing it with a random character, for example, "房" is replaced with "止"; and a 10% probability of keeping it unchanged, for example, "大" remains unchanged. This process continues until no more words can be masked or the number of masked characters exceeds 20% of the total number of characters in the merchant name. Then, the masked merchant name obtained through the above steps is used as the input of the merchant brand pre-training model, and its prediction target is the original character of each masked character.

[0096] Therefore, in the embodiments of the present application, by introducing the merchant abbreviation, the first preset model can be enabled to pay more attention to the suffix information of the merchant name that is irrelevant to the abbreviation, and promote the model to learn the semantic relationship among the merchant name, merchant brand and merchant category.

[0097] In another example, the second sample set involved in the foregoing step 3501 can be obtained through the following steps 35015 to 35017.

[0098] Step 35015: Obtain a sample candidate brand set of the sample merchant corresponding to the sample merchant name.

[0099] Specifically, in the case that the sample brand information includes the sample brand name of the sample brand, the sample brand operation category and the sample brand identifier, and the sample brand name includes the sample brand abbreviation and the full sample brand name, this step may specifically include:

[0100] Obtain a sample brand abbreviation that matches the sample merchant abbreviation according to the sample merchant abbreviation of the sample merchant;

[0101] When the sample merchant operation category of the sample merchant matches the sample brand operation category, add the sample brand name and the sample brand identifier to the initial sample candidate brand set corresponding to the sample merchant, so as to obtain the sample candidate brand set of the sample merchant.

[0102] Step 35016: According to the random interference algorithm, construct positive name pairs and negative name pairs of sample merchant names and sample brand names based on the sample candidate brand set, wherein the random interference algorithm is determined by the number of pieces of sample candidate brand information in the sample candidate brand set.

[0103] Step 35017: Determine the positive name pairs and negative name pairs as the second sample set.

[0104] Furthermore, in one instance, the sample candidate brand set does not include sample candidate brand information. Based on this, step 35016 may specifically include:

[0105] In the case where the positive example name pair includes the first positive example name pair, the negative example name pair includes the first negative example name pair, and the first negative example name pair includes the first sub-negative example name pair and the second sub-negative example name pair, a fourth preset number of first positive example name pairs are generated based on the sample merchant name and the empty sample candidate brand name.

[0106] Based on the sample merchant name and the first random sample brand name, a fifth preset number of first sub-negative example name pairs are generated; and based on the random sample merchant name and the empty sample brand name, a sixth preset number of second sub-negative example name pairs are generated; wherein,

[0107] The random sample merchant name is obtained by replacing the sample merchant name with the preset random sample brand abbreviation.

[0108] For example, positive and negative examples are constructed using the candidate brand set of sample merchants in the weakly supervised merchant dataset, enabling the second preset model to learn the semantic relationship between brand names and merchant names. Unlike current methods that directly and randomly replace brands to construct negative examples, this embodiment introduces merchant abbreviations, distinguishes between the abbreviation and the suffix of the merchant name, and randomly replaces each part with a certain probability. This method can more effectively generate negative examples with higher relevance, thereby improving the accuracy of merchant brand matching. Specifically, if the sample candidate brand set is an empty set, meaning that the sample candidate brand set does not include sample candidate brand information, it can be kept unchanged with a 50% probability, which is a positive example merchant-brand pair without a brand, such as Vienna Hotel-None; with a 25% probability, a random brand is added, such as Vienna Hotel-SF Express; with a 25% probability, a brand is randomly obtained, and the abbreviation of this brand is replaced with the abbreviation of the current merchant to generate a new incorrect brand, such as Vienna Hotel being replaced with Yonghui Hotel, and then the second sub-negative example name pair is obtained, namely Yonghui Hotel-None. Thus, the above latter method can all obtain the first negative example name pair.

[0109] In another instance, the sample candidate brand set includes sample candidate brand information. Based on this, step 35016 may specifically include:

[0110] In the case where the positive example name pair includes the second positive example name pair, the negative example name pair includes the second negative example name pair, the second negative example name pair includes the third sub-negative example name pair, the fourth sub-negative example name pair, the fifth sub-negative example name pair, and the sixth sub-negative example name pair, the sample candidate brand information includes the sample candidate brand name, the sample candidate brand name includes the sample candidate brand abbreviation, and the sample candidate brand abbreviation includes prefix and suffix words, the sample merchant name and the sample candidate brand name are determined as the second positive example name pair;

[0111] Based on the sample merchant name and the sample replacement brand name, generate a seventh preset number of third sub-negative example name pairs. The sample replacement brand name is obtained by replacing the second random sample brand name.

[0112] Based on the sample merchant name and the empty sample candidate brand name, generate the eighth preset number of fourth sub-negative example name pairs;

[0113] Based on the sample merchant name and the suffix replacement sample candidate brand name, generate the ninth preset number of fifth sub-negative example name pairs. The suffix replacement sample candidate brand name is obtained by replacing the suffix of the sample candidate brand abbreviation with a preset random suffix.

[0114] Based on the sample merchant name and the sample replacement brand abbreviation, generate the tenth preset number of sixth sub-negative sample name pairs. The sample replacement brand abbreviation is obtained by replacing the sample candidate brand name with a preset random abbreviation.

[0115] For example, such as Figure 5 As shown, if the sample candidate brand set is non-empty, meaning it includes sample candidate brand information, it is kept unchanged with a 60% probability, resulting in a positive merchant-brand pair containing the brand, such as Vienna Hotel-Vienna Hotel; it is replaced with a random brand with an 8% probability, such as Vienna Hotel-Yonghui Supermarket; it is deleted with a 12% probability, such as Vienna Hotel-None; it is replaced with the suffix of the current brand abbreviation with the suffix of a random brand abbreviation with a 12% probability, generating a new incorrect brand, such as replacing "Hotel" in Vienna Hotel with "Old Hot Pot" in Xiaolongkan Hot Pot, resulting in Vienna Hotel-Vienna Hot Pot; it is replaced with the abbreviation of the current brand with a random brand abbreviation with an 8% probability, generating a new incorrect brand, such as replacing "Vienna" in Vienna Hotel with "SF Express" in SF Hotel, resulting in Vienna Hotel-SF Express Hotel. All of the above methods can obtain a second negative name pair. These second positive and second negative name pairs will be used as input to the merchant brand pre-training model, and the prediction target result is whether the input merchant and its brand pair match.

[0116] Thus, by selectively replacing merchant abbreviations and suffixes and constructing negative examples, more relevant negative example brands can be generated more effectively, thereby improving the accuracy of matching merchants with their brands. Furthermore, this embodiment can effectively prevent erroneous associations by introducing merchant business categories and constructing negative example brands. Introducing merchant business categories can also distinguish merchant category information, preventing associations between irrelevant merchants and brands. Additionally, this embodiment can further improve the accuracy of merchant brand associations by constructing negative examples of merchant brands in the merchant brand pre-training model, enabling the model to effectively learn the semantic relationships between keywords related to merchant categories.

[0117] Secondly, regarding step 3502, based on the above example, the merchant pre-training task involves training the merchant brand pre-training model. In the input part of the preset model, this embodiment introduces special symbols "[CLS]" and "[SEP]", which are used respectively for brand association determination placeholders and to separate different parts of the input.

[0118] In one example, for the first sample set constructed above, the model input sequence for the merchant masking language model task is a concatenation of the following text parts: [CLS] + full name of the merchant after masking + [SEP] + merchant address + [SEP] + merchant business category + [SEP] + merchant brand; for the merchant brand determination task, the model input is a concatenation of the following text parts: [CLS] + full name of the merchant + [SEP] + merchant address + [SEP] + merchant business category + [SEP] + processed brand.

[0119] In another example, for the second sample set constructed above, for the merchant mask language model task, the model outputs the probability predicted for each character in the dictionary at each mask character position, thereby training the model to learn the original character of each mask character; for the merchant brand determination task, the model outputs the probability of merchant-brand association at the [CLS] symbol position, thereby training the model to learn whether the input merchant-brand pair matches.

[0120] Then, the two first-decision modules and the second-decision module are trained alternately in batches, using cross-entropy as the error and Adam as the optimizer for model optimization. Each training iteration trains on the full dataset of both task training sets, and training stops after reaching the maximum number of iterations, resulting in the merchant brand pre-trained model.

[0121] Thus, based on the merchant brand association determination model obtained above, step 340 may specifically include:

[0122] Step 3401: Calculate the association probability value between the merchant name and each candidate brand using the merchant brand association determination model.

[0123] Step 3402: Filter the target association probability value from the association probability values. The target association probability value meets the preset association conditions.

[0124] Step 3403: The candidate brand information corresponding to the probability value associated with the target is determined as the target brand information.

[0125] Based on this, in one or more possible embodiments, the merchant information also includes the merchant address. Thus, step 3401 above may specifically include:

[0126] Merchant name, merchant address, merchant business category, and candidate brand information are input into the merchant brand association determination model. The association probability value between the merchant name and each candidate brand information is calculated by the merchant association determination model. The association probability value is used to characterize the semantic association degree between the merchant name, merchant business category, and candidate brand information.

[0127] In another or more possible embodiments, the preset association conditions include a preset average value and a first preset condition. Based on this, step 3402 above may specifically include:

[0128] Filter out at least one first association probability value that is greater than or equal to a preset tie value from the association probability values;

[0129] The first association probability value that satisfies the first preset condition among at least one first association probability value is determined as the target association probability value.

[0130] For example, the merchant name, merchant address, and each candidate brand name are used to construct input sequences. These sequences are then input into the merchant brand association determination model to obtain the association probability value between the merchant name and each candidate brand. If no association score is higher than the first association probability value of 0.5, it means that the merchant has no associated brands. If only one association score is higher than 0.5, then this brand is the associated brand of the merchant. If multiple association scores are higher than 0.5, then the brand with the highest score is taken as the associated brand of the merchant.

[0131] In summary, the process involves acquiring merchant information, including the merchant name and business category. Then, based on the merchant name, it retrieves matching brand names. If the business category and brand name match the corresponding brand business category, it generates a candidate brand set for the merchant, containing candidate brand information for at least one candidate brand. Finally, a merchant brand association determination model identifies the target brand information corresponding to the merchant from the candidate brand set. By incorporating merchant abbreviation and category information into the merchant-brand matching process, the probability of missed or incorrect merchant associations caused by full-word matching of merchant and brand names can be effectively reduced, improving the accuracy of merchant-brand association and ultimately enabling effective merchant management.

[0132] Furthermore, by introducing merchant business categories and negative example brand construction, erroneous associations can be effectively prevented, and associations between unrelated merchants and brands can be prevented. This allows the model to effectively learn the semantic relationships of keywords related to merchant categories, thereby further preventing incorrect associations between merchants and their brands. Similarly, by selectively replacing merchant abbreviations and suffixes and constructing negative examples, more relevant negative example brands can be generated more effectively, thereby improving the accuracy of merchant brand matching.

[0133] In addition, the embodiments of this application do not require targeted crawling of keywords in each brand platform webpage, nor do they require obtaining the keywords and non-keywords of each brand. It is only necessary to construct a dataset containing merchant and brand names, as well as the business categories of merchants and brands, to complete the association between merchants and their brands. In this way, the requirements for brand information are lower and easier to obtain, reducing the cost and complexity of the process of associating merchants with their brands.

[0134] Based on the same inventive concept, this application also provides an information processing device. (Specifically combined with...) Figure 7 A detailed explanation will be provided.

[0135] Figure 7 This is a schematic diagram of the structure of an information processing device provided in one embodiment of this application.

[0136] In some embodiments of this application, Figure 7 The information processing device shown can be set up in, for example Figure 1 In the information processing architecture shown.

[0137] like Figure 7 As shown, the information processing device 70 may specifically include:

[0138] The acquisition module 701 is used to acquire merchant information, which includes the merchant name and the merchant's business category.

[0139] Processing module 702 is used to obtain the brand name that matches the merchant name, and the brand name corresponds to the brand's business category.

[0140] The generation module 703 is used to generate a set of candidate brands corresponding to the merchant based on the brand name when the merchant's business category matches the brand's business category. The set of candidate brands includes candidate brand information of at least one candidate brand.

[0141] The determination module 704 is used to determine the target brand information corresponding to the merchant in the candidate brand set through the merchant brand association determination model. The target brand information includes the brand name corresponding to the merchant name.

[0142] In this embodiment, merchant abbreviation information and merchant category information can be introduced into the merchant-brand matching process. This can effectively reduce the probability of missed or incorrect merchant associations caused by full-word matching of merchant and brand names, improve the accuracy of merchant-brand association, and thus achieve effective merchant management. Furthermore, this embodiment does not require targeted crawling of keywords from each brand platform webpage, nor does it require obtaining keywords and non-keywords for each brand. It only needs to construct a dataset containing merchant and brand names, as well as merchant and brand business categories, to complete the association between merchants and brands. This reduces the requirements for brand information, makes it easier to obtain, and lowers the cost and complexity of the merchant-brand association process.

[0143] The information processing device 70 in the embodiments of this application will be described in detail below.

[0144] In one or more optional embodiments, the acquisition module 701 in this application embodiment can be specifically used to match the merchant abbreviation with the brand abbreviation in the preset brand information set when the merchant name includes the merchant abbreviation, and obtain the matching result;

[0145] If the matching results indicate that there is a brand abbreviation corresponding to the merchant abbreviation in the preset brand information set, the brand information of the brand abbreviation corresponding to the merchant abbreviation is obtained. The brand information includes the brand name and brand business category.

[0146] In another or more alternative embodiments, the generation module 703 in this application embodiment can be specifically used to obtain an initial candidate brand set corresponding to the merchant when the brand information also includes the brand identity identifier of the brand and the merchant's business category matches the brand's business category.

[0147] Add the brand name and brand identity to the initial candidate brand set to obtain the candidate brand set corresponding to the merchant.

[0148] In another or more alternative embodiments, the determining module 704 in this application embodiment can be specifically used to calculate the association probability value between the merchant name and each candidate brand through the merchant brand association determination model.

[0149] Filter the target association probability values ​​from the association probability values, and the target association probability values ​​meet the preset association conditions;

[0150] The candidate brand information corresponding to the probability value associated with the target is determined as the target brand information.

[0151] In another or more alternative embodiments, the determining module 704 in this application embodiment can be specifically used to input the merchant name, merchant address, merchant business category and candidate brand information into the merchant brand association determination model when the merchant information also includes the merchant address, and calculate the association probability value between the merchant name and each candidate brand information through the merchant association determination model. The association probability value is used to characterize the semantic association degree between the merchant name, the merchant business category and the candidate brand information.

[0152] In another or more alternative embodiments, the determining module 704 in this application embodiment can be specifically used to filter out at least one first association probability value that is greater than or equal to a preset average value from the association probability values ​​when the preset association conditions include a preset average value and a first preset condition.

[0153] The first association probability value that satisfies the first preset condition among at least one first association probability value is determined as the target association probability value.

[0154] In another or more alternative embodiments, the information processing device 70 in this application embodiment may further include a training module; wherein,

[0155] The acquisition module 701 in this embodiment can also be used to acquire a sample set, which includes a first sample set and a second sample set. The first sample set includes information used to characterize the degree of semantic association between the sample merchant name, the sample merchant business category and the sample brand information. The second sample set includes information used to characterize the degree of semantic association between the sample merchant name and the sample brand information.

[0156] The training module is used to train a first preset model based on a first sample set, and to obtain an initial first decision sub-model when the first preset model meets the first preset training conditions; and to train a second preset model based on a second sample set, and to obtain an initial second decision sub-model when the second preset model meets the second preset training conditions.

[0157] The generation module 703 in this embodiment can also be used to generate a merchant brand pre-trained model based on the initial first decision sub-model and the initial second decision sub-model.

[0158] The acquisition module 701 in this embodiment can also be used to acquire a manually labeled set of artificial samples, which includes multiple artificial association information, including a relationship pair between artificial business information and artificial brand information.

[0159] The training module can also be used to train a merchant brand pre-trained model based on multiple manually associated information, and to obtain a merchant brand association determination model when the merchant brand training model meets the third preset training condition.

[0160] In another or more alternative embodiments, the information processing device 70 in this application embodiment may further include a word segmentation module, a tagging module, and a masking module; wherein,

[0161] The word segmentation module is used to segment the full name of the sample merchant when the sample merchant name includes the full name of the sample merchant, and obtain the full name word sequence. The full name word sequence includes N words, where N is a positive integer.

[0162] The tagging module is used to tag each word in N words using a preset tagging algorithm to obtain a tagging sequence. The tagging sequence includes a first identifier and a second identifier. The first identifier is used to identify the first word in the N words that can be masked, and the second identifier is used to identify the second word in the N words that cannot be masked.

[0163] The masking module is used to mask the first word using a preset masking algorithm to obtain the masked name of the sample merchant.

[0164] The determining module 704 in this embodiment can also be used to determine the sample merchant mask name, sample merchant name, sample merchant identity identifier, sample merchant business category and sample brand name as the first sample set, and the sample merchant identity identifier is the identity identifier of the sample merchant corresponding to the sample merchant name.

[0165] In another or more alternative embodiments, the sample merchant name may further include j sample merchant abbreviations, which are determined by the text in the full name of the sample merchant;

[0166] The first word is the word following the first merchant abbreviation among the j sample merchant abbreviations, and which does not belong to the j sample merchant abbreviations, where j is a positive integer.

[0167] In another or more alternative embodiments, the masking module in this application embodiment can be specifically used to replace a first preset number of characters in the first word with preset characters and replace a second preset number of characters in the first word with random characters when the first word includes i characters, i being a positive integer greater than 1.

[0168] According to the arrangement order of i characters, the third preset number of original characters, preset characters and random characters in the i characters are concatenated to obtain the sample merchant mask name.

[0169] It should be noted that the first preset training conditions include at least one of the following: the number of iterations of the first decision sub-model is greater than or equal to the first preset number of iterations, and the decision result output by the first decision sub-model includes the first word of the sample merchant mask name before it is masked.

[0170] In another or more alternative embodiments, the information processing device 70 in this application embodiment may further include a construction module; wherein,

[0171] The acquisition module 701 in this embodiment can also be used to acquire a set of sample candidate brands corresponding to the sample merchant name.

[0172] The construction module is used to construct positive and negative name pairs of sample merchant names and sample brand names according to the sample candidate brand set, based on the random interference algorithm. The random interference algorithm is determined by the number of sample candidate brand information in the sample candidate brand set.

[0173] The determination module 704 in this embodiment can also be used to determine the positive example name pairs and negative example name pairs as the second sample set.

[0174] In another or more alternative embodiments, the acquisition module 701 in this application embodiment can also be used to, when the sample brand information includes the sample brand name, sample brand business category and sample brand identity identifier, and the sample brand name includes the sample brand abbreviation and the full name of the sample brand, obtain the sample brand abbreviation that matches the sample merchant abbreviation based on the sample merchant abbreviation.

[0175] If the business category of a sample merchant matches the business category of a sample brand, the sample brand name and sample brand identity identifier are added to the initial candidate brand set corresponding to the sample merchant, thus obtaining the sample candidate brand set of the sample merchant.

[0176] In another or more alternative embodiments, the construction module in the present application embodiment can be specifically used to generate a fourth preset number of first positive example name pairs based on the sample merchant name and empty sample candidate brand name when the positive example name pair includes a first positive example name pair, the negative example name pair includes a first negative example name pair, the first negative example name pair includes a first sub-negative example name pair and a second sub-negative example name pair, and the sample candidate brand set does not include sample candidate brand information;

[0177] Based on the sample merchant name and the first random sample brand name, a fifth preset number of first sub-negative example name pairs are generated; and based on the random sample merchant name and the empty sample brand name, a sixth preset number of second sub-negative example name pairs are generated; wherein,

[0178] The random sample merchant name is obtained by replacing the sample merchant name with the preset random sample brand abbreviation.

[0179] In another or more alternative embodiments, the construction module in this application embodiment can be specifically used to determine the sample merchant name and the sample candidate brand name as the second positive example name pair when the positive example name pair includes the second positive example name pair, the negative example name pair includes the second negative example name pair, the second negative example name pair includes the third sub-negative example name pair, the fourth sub-negative example name pair, the fifth sub-negative example name pair and the sixth sub-negative example name pair, the sample candidate brand information includes the sample candidate brand name, the sample candidate brand name includes the sample candidate brand abbreviation, and the sample candidate brand abbreviation includes prefix and suffix words, and when the sample candidate brand set includes the sample candidate brand information;

[0180] Based on the sample merchant name and the sample replacement brand name, generate a seventh preset number of third sub-negative example name pairs. The sample replacement brand name is obtained by replacing the second random sample brand name.

[0181] Based on the sample merchant name and the empty sample candidate brand name, generate the eighth preset number of fourth sub-negative example name pairs;

[0182] Based on the sample merchant name and the suffix replacement sample candidate brand name, generate the ninth preset number of fifth sub-negative example name pairs. The suffix replacement sample candidate brand name is obtained by replacing the suffix of the sample candidate brand abbreviation with a preset random suffix.

[0183] Based on the sample merchant name and the sample replacement brand abbreviation, generate the tenth preset number of sixth sub-negative sample name pairs. The sample replacement brand abbreviation is obtained by replacing the sample candidate brand name with a preset random abbreviation.

[0184] Based on the same inventive concept, this application also provides an information processing device. (Specifically combined with...) Figure 8 A detailed explanation will be provided.

[0185] Figure 8 This is a schematic diagram of the structure of an information processing device provided in one embodiment of this application.

[0186] like Figure 8 As shown, the information processing device may include at least one of the following as described in the embodiments of this application: an electronic device, a server. The information processing device may include a processor 801 and a memory 802 storing computer program instructions.

[0187] Specifically, the processor 801 may include a central processing unit (CPU), an application-specific integrated circuit (ASIC), or one or more integrated circuits that can be configured to implement the embodiments of this application.

[0188] Memory 802 may include mass storage for data or instructions. For example, and not limitingly, memory 802 may include a hard disk drive (HDD), floppy disk drive, flash memory, optical disk, magneto-optical disk, magnetic tape, or Universal Serial Bus (USB) drive, or a combination of two or more of these. Where appropriate, memory 802 may include removable or non-removable (or fixed) media. Where appropriate, memory 802 may be internal or external to the integrated gateway disaster recovery device. In a particular embodiment, memory 802 is non-volatile solid-state memory. In a particular embodiment, memory 802 includes solid-state storage (ROM). Where appropriate, the ROM may be a mask-programmed ROM, a programmable ROM (PROM), an erasable PROM (EPROM), an electrically erasable PROM (EEPROM), an electrically rewritable ROM (EAROM), or flash memory, or a combination of two or more of these.

[0189] The processor 801 implements any of the information processing methods described in the above embodiments by reading and executing computer program instructions stored in the memory 802.

[0190] In one example, the information processing device may further include a communication interface 803 and a bus 810. Wherein, as... Figure 8 As shown, the processor 801, memory 802, and communication interface 803 are connected through bus 810 and complete communication with each other.

[0191] The communication interface 803 is mainly used to realize communication between various modules, devices, units and / or equipment in the embodiments of this application.

[0192] Bus 810 includes hardware, software, or both, that couples components of a flow control device together. For example, and not limitingly, the bus may include an Accelerated Graphics Port (AGP) or other graphics bus, an Enhanced Industry Standard Architecture (EISA) bus, a Front Side Bus (FSB), HyperTransport (HT) interconnect, an Industry Standard Architecture (ISA) bus, an Infinite Bandwidth Interconnect, a Low Pin Count (LPC) bus, a memory bus, a Microchannel Architecture (MCA) bus, a Peripheral Component Interconnect (PCI) bus, a PCI-Express (PCI-X) bus, a Serial Advanced Technology Attachment (SATA) bus, a Video Electronics Standards Association Local (VLB) bus, or other suitable buses, or combinations of two or more of these. Where appropriate, bus 810 may include one or more buses. Although specific buses are described and illustrated in embodiments of this application, any suitable bus or interconnect is contemplated herein.

[0193] The data processing device can execute the information processing method described in the embodiments of this application, thereby achieving the combination Figures 1 to 6 The described information processing methods and apparatus.

[0194] Furthermore, in conjunction with the information processing methods in the above embodiments, this application embodiment can provide a computer-readable storage medium for implementation. This computer-readable storage medium stores computer program instructions; when these computer program instructions are executed by a processor, they implement any of the information processing methods in the above embodiments.

[0195] It should be clarified that this application is not limited to the specific configurations and processes described above and shown in the figures. For the sake of brevity, detailed descriptions of known methods are omitted here. In the above embodiments, several specific steps are described and shown as examples. However, the method process of this application is not limited to the specific steps described and shown. Those skilled in the art can make various changes, modifications, and additions, or change the order of steps, after understanding the spirit of this application.

[0196] The functional blocks shown in the above block diagram can be implemented as hardware, software, firmware, or a combination thereof. When implemented in hardware, they can be, for example, electronic circuits, application-specific integrated circuits (ASICs), appropriate firmware, plug-ins, function cards, etc. When implemented in software, the elements of this application are programs or code segments used to perform the required tasks. Programs or code segments can be stored on a machine-readable medium or transmitted over a transmission medium or communication link via data signals carried on a carrier wave. "Machine-readable medium" can include any medium capable of storing or transmitting information. Examples of machine-readable media include electronic circuits, semiconductor memory devices, ROM, flash memory, erasable ROM (EROM), floppy disks, CD-ROMs, optical disks, hard disks, fiber optic media, radio frequency (RF) links, etc. Code segments can be downloaded via computer networks such as the Internet, intranets, etc.

[0197] It should also be noted that the exemplary embodiments mentioned in this application describe methods or systems based on a series of steps or apparatus. However, this application is not limited to the order of the above steps; that is, the steps can be performed in the order mentioned in the embodiments, or in a different order, or several steps can be performed simultaneously.

[0198] The above are merely specific embodiments of this application. Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, modules, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here. It should be understood that the protection scope of this application is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in this application, and these modifications or substitutions should all be covered within the protection scope of this application.

Claims

1. An information processing method, comprising: Obtain merchant information, including merchant name and merchant business category; Based on the merchant name, obtain the brand name that matches the merchant name, and the brand name corresponds to the brand's business category; If the merchant's business category matches the brand's business category, a set of candidate brands corresponding to the merchant is generated based on the brand name. The set of candidate brands includes candidate brand information for at least one candidate brand. Obtain a sample set, which includes a first sample set and a second sample set. The first sample set includes information used to characterize the semantic association between the sample merchant name, the sample merchant business category and the sample brand information. The second sample set includes information used to characterize the semantic association between the sample merchant name and the sample brand information. The first preset model is trained based on the first sample set, and an initial first judgment sub-model is obtained when the first preset model meets the first preset training conditions. Furthermore, the second preset model is trained based on the second sample set, and an initial second decision sub-model is obtained when the second preset model meets the second preset training conditions. Based on the initial first decision sub-model and the initial second decision sub-model, a merchant brand pre-training model is generated; a manually labeled sample set is obtained, the sample set including multiple manually associated information, the manually associated information including relationship pairs between manually labeled merchant information and manually labeled brand information; The merchant brand pre-training model is trained based on the multiple artificial association information. When the merchant brand pre-training model meets the third preset training condition, a merchant brand association determination model is obtained. The merchant brand association determination model is used to determine the target brand information corresponding to the merchant in the candidate brand set. The target brand information includes the brand name corresponding to the merchant name.

2. The method according to claim 1, characterized in that, The merchant name includes a merchant abbreviation; obtaining the brand name matching the merchant name includes: The merchant abbreviation is matched with the brand abbreviations in the preset brand information set to obtain the matching results; If the matching result indicates that there is a brand abbreviation corresponding to the merchant abbreviation in the preset brand information set, the brand information of the brand abbreviation corresponding to the merchant abbreviation is obtained, and the brand information includes the brand name and brand business category.

3. The method according to claim 2, characterized in that, The brand information also includes the brand identity identifier; when the merchant's business category matches the brand's business category, generating a set of candidate brands corresponding to the merchant based on the brand name includes: If the merchant's business category matches the brand's business category, obtain an initial set of candidate brands corresponding to the merchant; The brand name and the brand identity identifier are added to the initial candidate brand set to obtain the candidate brand set corresponding to the merchant.

4. The method according to claim 1, characterized in that, The step of determining the target brand information corresponding to the merchant from the candidate brand set through the merchant brand association determination model includes: Using the merchant brand association determination model, the association probability value between the merchant name and each candidate brand is calculated respectively. Filter target association probability values ​​from the association probability values, wherein the target association probability values ​​satisfy preset association conditions; The candidate brand information corresponding to the probability value associated with the target is determined as the target brand information.

5. The method according to claim 4, characterized in that, The merchant information also includes the merchant address; The step of calculating the association probability value between the merchant name and each candidate brand using the merchant brand association determination model includes: The merchant name, merchant address, merchant business category, and candidate brand information are input into the merchant brand association determination model. The association probability value between the merchant name and each candidate brand information is calculated through the merchant association determination model. The association probability value is used to characterize the semantic association degree between the merchant name, the merchant business category, and the candidate brand information.

6. The method according to claim 4 or 5, characterized in that, The preset association conditions include a preset average value and a first preset condition; The step of filtering the target association probability value from the association probability values ​​includes: Filter out at least one first association probability value that is greater than or equal to the preset average value from the association probability values; The first association probability value that satisfies the first preset condition among the at least one first association probability value is determined as the target association probability value.

7. The method according to claim 1, characterized in that, The sample merchant name includes the full name of the sample merchant; obtaining the sample set includes: The full name of the sample merchant is segmented into words to obtain a sample full name word sequence, which includes N words, where N is a positive integer. A preset labeling algorithm is used to label each of the N words to obtain a labeling sequence. The labeling sequence includes a first identifier and a second identifier. The first identifier is used to identify the first word among the N words that can be masked, and the second identifier is used to identify the second word among the N words that cannot be masked. The first word is masked using a preset masking algorithm to obtain the masked name of the sample merchant. The sample merchant mask name, the sample merchant name, the sample merchant identity identifier, the sample merchant business category, and the sample brand name are determined as the first sample set, and the sample merchant identity identifier is the identity identifier of the sample merchant corresponding to the sample merchant name.

8. The method according to claim 7, characterized in that, The sample merchant name also includes j sample merchant abbreviations, which are determined by the text in the full name of the sample merchant; The first word is the word following the first merchant abbreviation among the j sample merchant abbreviations, and which does not belong to the j sample merchant abbreviations, where j is a positive integer.

9. The method according to claim 7, characterized in that, The first word comprises i characters, where i is a positive integer greater than 1; the step of masking the first word using a preset masking algorithm to obtain the sample merchant mask name includes: Replace a first preset number of characters in the i characters with preset characters, and replace a second preset number of characters in the i characters with random characters; According to the arrangement order of the i characters, the third preset number of original characters, the preset characters, and the random characters in the i characters are concatenated to obtain the sample merchant mask name.

10. The method according to claim 1, characterized in that, The first preset training condition includes at least one of the following: the number of iterations of the first decision sub-model is greater than or equal to the first preset number of iterations; the decision result output by the first decision sub-model includes the first word of the sample merchant mask name before it is masked.

11. The method according to claim 1, characterized in that, The acquisition of the sample set includes: Obtain the sample candidate brand set of the sample merchants corresponding to the sample merchant names; According to the random interference algorithm, based on the sample candidate brand set, positive and negative example name pairs of sample merchant names and sample brand names are constructed. The random interference algorithm is determined by the number of sample candidate brand information in the sample candidate brand set. The positive example name pairs and the negative example name pairs are determined as the second sample set.

12. The method according to claim 11, characterized in that, The sample brand information includes the sample brand name, sample brand business category, and sample brand identity identifier. The sample brand name includes the sample brand abbreviation and the full name of the sample brand. The step of obtaining the sample candidate brand set of sample merchants corresponding to the sample merchant names includes: Based on the sample merchant's abbreviation, obtain the sample brand abbreviation that matches the sample merchant's abbreviation; If the business category of the sample merchant matches the business category of the sample brand, the sample brand name and the sample brand identity identifier are added to the initial candidate brand set corresponding to the sample merchant to obtain the sample candidate brand set of the sample merchant.

13. The method according to claim 11, characterized in that, The positive example name pair includes a first positive example name pair, the negative example name pair includes a first negative example name pair, and the first negative example name pair includes a first sub-negative example name pair and a second sub-negative example name pair; The step of constructing positive and negative example name pairs of sample merchant names and sample brand names based on the sample candidate brand set using a random interference algorithm includes: If the sample candidate brand information is not included in the sample candidate brand set, a fourth preset number of the first positive example name pairs are generated based on the sample merchant name and the empty sample candidate brand name. Based on the sample merchant name and the first random sample brand name, a fifth preset number of first sub-negative example name pairs are generated; and based on the random sample merchant name and the empty sample candidate brand name, a sixth preset number of second sub-negative example name pairs are generated; wherein, The random sample merchant name is obtained by replacing the sample merchant name with a preset random sample brand abbreviation.

14. The method according to claim 11, characterized in that, The positive example name pair includes a second positive example name pair, the negative example name pair includes a second negative example name pair, the second negative example name pair includes a third sub-negative example name pair, a fourth sub-negative example name pair, a fifth sub-negative example name pair and a sixth sub-negative example name pair, the sample candidate brand information includes a sample candidate brand name, the sample candidate brand name includes a sample candidate brand abbreviation, and the sample candidate brand abbreviation includes a prefix and a suffix. The step of constructing positive and negative example name pairs of sample merchant names and sample brand names based on the sample candidate brand set using a random interference algorithm includes: If the sample candidate brand set includes sample candidate brand information, the sample merchant name and the sample candidate brand name are determined as the second positive example name pair; Based on the sample merchant name and the sample replacement brand name, a seventh preset number of third sub-negative example name pairs are generated, wherein the sample replacement brand name is obtained by replacing the second random sample brand name. Based on the sample merchant names and empty sample candidate brand names, generate an eighth preset number of fourth sub-negative example name pairs; Based on the sample merchant name and the suffix replacement sample candidate brand name, a ninth preset number of fifth sub-negative example name pairs are generated. The suffix replacement sample candidate brand name is obtained by replacing the suffix of the sample candidate brand abbreviation with a preset random suffix. Based on the sample merchant name and the sample replacement brand abbreviation, a tenth preset number of sixth sub-negative example name pairs are generated. The sample replacement brand abbreviation is obtained by replacing the sample candidate brand name with a preset random abbreviation.

15. An information processing apparatus, the apparatus comprising: The acquisition module is used to acquire merchant information, which includes the merchant name and the merchant's business category. The processing module is used to obtain a brand name that matches the merchant name, and the brand name corresponds to the brand's business category. The generation module is used to generate a set of candidate brands corresponding to the merchant based on the brand name when the merchant's business category matches the brand's business category. The set of candidate brands includes candidate brand information of at least one candidate brand. The acquisition module is used to acquire a sample set, which includes a first sample set and a second sample set. The first sample set includes information used to characterize the semantic association between the sample merchant name, the sample merchant business category and the sample brand information. The second sample set includes information used to characterize the semantic association between the sample merchant name and the sample brand information. The training module is used to train the first preset model based on the first sample set, and to obtain an initial first judgment sub-model when the first preset model meets the first preset training conditions. Furthermore, the second preset model is trained based on the second sample set, and an initial second decision sub-model is obtained when the second preset model meets the second preset training conditions. Based on the initial first decision sub-model and the initial second decision sub-model, a merchant brand pre-training model is generated; a manually labeled sample set is obtained, the sample set including multiple manually associated information, the manually associated information including relationship pairs between manually labeled merchant information and manually labeled brand information; The merchant brand pre-training model is trained based on the multiple artificial association information. When the merchant brand pre-training model meets the third preset training condition, a merchant brand association determination model is obtained. The determination module is used to determine the target brand information corresponding to the merchant in the candidate brand set through the merchant brand association determination model. The target brand information includes the brand name corresponding to the merchant name.

16. A computing device, the device comprising: Processor and memory storing computer program instructions; When the processor executes the computer program instructions, it implements the information processing method as described in any one of claims 1-14.

17. A storage medium storing computer program instructions, which, when executed by a processor, implement the information processing method as described in any one of claims 1-14.

Images