Method and device for determining optimization direction of merchant name translation model

Abstract

The purpose of this application is to provide a method and device for determining the optimization direction of a merchant name translation model. Compared with the prior art, this application obtains the translation results of the merchant name translation model for multiple merchant information; determines the quantitative evaluation index values of each translation result based on multiple preset quantitative evaluation indicators; determines the evaluation result corresponding to each translation result according to the quantitative evaluation index value through preset quantitative evaluation rules; and determines the optimization direction of the merchant name translation model corresponding to the evaluation result. In this way, the optimization direction of the merchant name translation model is determined, so that relevant technical personnel can determine the specific optimization content based on the optimization direction, which greatly reduces the time and workload required for manual investigation and saves the optimization cost of the merchant name translation model.

Description

Technical Field

[0001] The present application relates to the field of computer technology, and in particular to a technology for determining an optimization direction of a merchant name translation model. Background Art

[0002] The invention patent application with application number 202110426399.X discloses a method and device for translating merchant names. The application cleans the merchant name to be translated at the character level and word level to obtain a cleaned merchant name; identifies the address information, organizational form information and store information contained in the cleaned merchant name to obtain a first identified merchant name; identifies the industry information and corporate name information in the first identified merchant name to obtain a second identified merchant name; translates the second identified merchant name to obtain a translated merchant name.

[0003] Because the existing merchant name translation model lacks a method to automatically determine the optimization direction, the optimization direction of the merchant name translation model can only be determined through manual investigation. In actual application scenarios, the number of merchant names to be translated may be in the millions. In this case, manual investigation is extremely time-consuming and costly. Therefore, a method is needed to automatically determine the optimization direction, provide directional guidance for the specific optimization of subsequent technical personnel, narrow the scope of manual investigation, and save optimization costs. Summary of the Invention

[0004] The purpose of this application is to provide a method and device for determining the optimization direction of a merchant name translation model.

[0005] According to one aspect of the present application, a method for determining an optimization direction of a merchant name translation model is provided, wherein the method includes:

[0006] Obtain translation results of the merchant name translation model for multiple merchant information;

[0007] Determining the quantitative evaluation index values of each translation result based on a plurality of preset quantitative evaluation indexes;

[0008] Determining the evaluation result corresponding to each translation result according to the quantitative evaluation index value using a preset quantitative evaluation rule;

[0009] Determine an optimization direction of the merchant name translation model corresponding to the evaluation result.

[0010] Furthermore, the merchant name translation model includes a vocabulary, and before obtaining the translation results of the merchant name translation model for the plurality of merchant information, the method further includes:

[0011] Determining a vocabulary optimization direction of the merchant name translation model based on the merchant information, the vocabulary, and a preset filtering threshold;

[0012] The step of obtaining translation results of the merchant name translation model for the plurality of merchant information includes:

[0013] Obtain translation results of the merchant name translation model optimized according to the vocabulary optimization direction for multiple merchant information.

[0014] Furthermore, the determining of the vocabulary optimization direction of the merchant name translation model based on the merchant information, the vocabulary, and a preset filtering threshold includes:

[0015] Determining a field to be optimized based on the merchant information and the vocabulary;

[0016] Count the occurrence frequency of the field to be optimized in all merchant information;

[0017] Based on the occurrence frequency and a preset filtering threshold, a vocabulary optimization direction of the merchant name translation model is determined.

[0018] Optionally, the field to be optimized has a characteristic word, and determining the field to be optimized based on the merchant information and the word library includes:

[0019] Filtering out merchant information having the characteristic words from all merchant information;

[0020] The field in the merchant information that is located before the feature word, has irrelevant information deleted, and is not stored in the vocabulary is determined as the field to be optimized, wherein the irrelevant information includes symbols and punctuation marks.

[0021] Optionally, the field to be optimized does not have a feature word, and the determining the field to be optimized based on the merchant information and the vocabulary includes:

[0022] Segmenting the merchant information according to a segmentation algorithm to obtain multiple fields;

[0023] The field in the merchant information that does not have a feature word and is not stored in the word library is determined as the field to be optimized.

[0024] Furthermore, the determining of the vocabulary optimization direction of the merchant name translation model based on the occurrence frequency and the preset filtering threshold includes:

[0025] The fields to be optimized whose occurrence frequency exceeds a preset filtering threshold are used as the content to be expanded in the vocabulary.

[0026] Preferably, the quantitative evaluation indicators include at least one of the following:

[0027] Address length, company name length, industry length, number of industries, merchant information length, length of the literal translation result not truncated according to the translation length limit, length of the translation result truncated according to the translation length limit, number of translated words, and translation retention score, where the translation retention score is the ratio of the translation result length to the literal translation result length.

[0028] Furthermore, the evaluation result includes at least one of the following:

[0029] Missing address recognition, over-recognition of addresses, incorrect industry identification, and excessive industry truncation;

[0030] The optimization direction corresponding to the evaluation result includes at least one of the following:

[0031] Improve the address recognition rules, industry recognition rules, truncation rules of the merchant name translation model, and improve the vocabulary of the merchant name translation model.

[0032] According to another aspect of the present application, a computer-readable medium is provided, on which computer-readable instructions are stored. The computer-readable instructions can be executed by a processor to implement the operations of the aforementioned method.

[0033] According to another aspect of the present application, a device for determining a translation model optimization direction is provided, wherein the device includes:

[0034] one or more processors; and

[0035] A memory storing computer-readable instructions that, when executed, cause the processor to perform the operations of the above-described method.

[0036] Compared to existing technologies, this application obtains the translation results of the merchant name translation model for multiple merchant information; determines the quantitative evaluation index values for each translation result based on multiple preset quantitative evaluation indicators; determines the evaluation result corresponding to each translation result using preset quantitative evaluation rules based on the quantitative evaluation index values; and determines the optimization direction of the merchant name translation model corresponding to the evaluation result. This method determines the optimization direction of the merchant name translation model, allowing relevant technical personnel to determine specific optimization content based on this optimization direction, significantly reducing the time and workload required for manual investigation and saving the optimization cost of the merchant name translation model. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Other features, objects and advantages of the present invention will become more apparent upon reading the detailed description of non-limiting embodiments made with reference to the following drawings:

[0038] Figure 1 A flow chart showing a method for determining an optimization direction of a merchant name translation model according to one aspect of the present application is shown;

[0039] Figure 2 A flow chart of a method for determining an optimization direction of a merchant name translation model according to a preferred embodiment of the present application is shown.

[0040] The same or similar reference numerals in the drawings represent the same or similar components. DETAILED DESCRIPTION

[0041] The present invention is further described in detail below with reference to the accompanying drawings.

[0042] In a typical configuration of the present application, the terminal, the device of the service network and the trusted party all include one or more processors (CPUs), input / output interfaces, network interfaces and memories.

[0043] Memory may include non-permanent storage in a computer-readable medium, random access memory (RAM) and / or non-volatile memory in the form of read-only memory (ROM) or flash RAM. Memory is an example of a computer-readable medium.

[0044] Computer-readable media includes permanent and non-permanent, removable and non-removable media that can be implemented by any method or technology to store information. The information can be computer-readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disc read-only memory (CD-ROM), digital versatile disc (DVD) or other optical storage, magnetic cassettes, magnetic disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined herein, computer-readable media does not include non-transitory media such as modulated data signals and carrier waves.

[0045] In order to further illustrate the technical means adopted by this application and the effects achieved, the technical solution of this application is clearly and completely described below in combination with the accompanying drawings and preferred embodiments.

[0046] Figure 1 A method for determining an optimization direction of a merchant name translation model provided in one aspect of the present application is shown, wherein the method includes:

[0047] S11 obtains translation results of the merchant name translation model for multiple merchant information;

[0048] S12 determines the quantitative evaluation index value of each translation result based on a plurality of preset quantitative evaluation indexes;

[0049] S13 determines the evaluation result corresponding to each translation result according to the quantitative evaluation index value using a preset quantitative evaluation rule;

[0050] S14 determines the optimization direction of the merchant name translation model corresponding to the evaluation result.

[0051] In this embodiment, in step S11, translation results of the merchant name translation model for multiple merchant information are obtained.

[0052] In this step, the merchant information is the full name of the merchant that needs to be translated for cross-border transactions. It is understandable that the merchant information is usually the full name of the company, including fields such as corporate name information (such as Tong Ren Tang), address information, industry information, and organizational form information (branch, subsidiary, and head office). The merchant name translation model cleans and translates the merchant information to obtain a direct translation result (i.e., the result obtained by performing all technical content except truncation in accordance with the invention patent application with application number 202110426399.X), and truncates the direct translation result according to the translation length limit to obtain the translation result.

[0053] In one application scenario, the merchant information is the full name of the company in Chinese, and the translation result is a truncated English translation of the direct translation of the full name of the company. For ease of understanding, the following content illustrates the technical solution of this application based on this application scenario, but it should be clear that this is not a restrictive description of this application. Any scenario in which this application can be applied is within the scope of protection of this application. The application scenarios in which this application can be applied include but are not limited to: translating merchant information in English and other languages into Chinese, translating merchant information in Chinese into other languages, and applying this merchant translation model to translate information other than merchant information, and then applying the technical solution protected by this application to determine the optimization direction.

[0054] In step S12, each quantitative evaluation index value of each translation result is determined based on a plurality of preset quantitative evaluation indexes.

[0055] In this step, quantitative evaluation indicators include, but are not limited to: address length, company name length, industry length, number of industries, merchant information length, length of the literal translation result without truncation according to the translation length limit, length of the translation result truncated according to the translation length limit, number of translated words, and translation retention score. The translation retention score is the ratio of the translation result length to the literal translation result length. A larger translation retention score indicates that less valid information is discarded during the translation process.

[0056] Here, the quantitative evaluation index is used to quantify the translation result, and the quantitative evaluation index value is the specific value of the corresponding quantitative evaluation index obtained based on the merchant information and the translation process. The quantitative evaluation index value is the quantitative value of the translation result.

[0057] Continuing in this step, the address length, company name length, industry length, number of industries and merchant information length are determined based on the content after cleaning and identifying the merchant information in accordance with the invention patent application with application number 202110426399.X. Taking the Chinese-English translation of the merchant information "Shanghai Youaixin Construction and Decoration Co., Ltd." as an example, the content after cleaning and identification is "Shanghai; Youaixin; Construction; Decoration; Co., Ltd.", then the address length is 2, the company name length is 3, the industry length is 4, the number of industries is 2, and the merchant information length is 13. Here, the content of the length statistics is the number of Chinese characters. Of course, the length can also be the length of the string in the storage space, etc.

[0058] It should be clear that since the address field usually does not need to be translated (for example, when translating from Chinese to English, it is only necessary to convert the address into pinyin, and it does not need to be translated into English), therefore, the direct translation result and the translation result do not include the address field, and thus the direct translation result of "Shanghai; Loving; Building; Decoration; Co., Ltd." is "YouaixinBuildingDecorationCoLtd" (where "building", "decoration", and "Co., Ltd." all have corresponding translation results stored in the vocabulary), and the length of the direct translation result is 31. When the length limit is 25, after truncation according to the truncation rules of the merchant information translation model, the translation result obtained is "YouaixinBuildingCoLtd", the length of the translation result is 21, the number of translated words is 2, and the translation retention score is 0.68. Here, the length statistics are the number of English letters. Of course, the length can also be the length of the string in the storage space, etc.

[0059] In step S13, the evaluation result corresponding to each translation result is determined according to the quantitative evaluation index value by using preset quantitative evaluation rules.

[0060] Here, the quantitative evaluation rule is one or more comprehensive thresholds set by combining multiple quantitative evaluation indicators. The evaluation results include but are not limited to omissions in address identification, excessive address identification, incorrect industry identification, excessive industry truncation, etc.

[0061] Continuing in this step, the quantitative evaluation rule can be understood as a decision tree for quantitatively evaluating the translation process and translation results. Taking an actual application scenario as an example, when the translation result is limited to 25 characters (taking the translation result in English as an example, that is, the total number of letters in the translation result should be less than 25), the quantitative evaluation rule and its representation include (the following content uses the mathematical symbols "∪" and "∩" to represent the word relationship "or" and "and"):

[0062] Rule 1:

[0063]

[0064] When the quantitative evaluation index values involved meet this rule, it means that the merchant name translation model may have missed identifying some address fields in the merchant information, resulting in the unrecognized address fields being mistaken for the company name fields, causing the company name fields to be too long, and thus resulting in a low translation retention score.

[0065] The evaluation result corresponding to the translation result that meets this rule is address recognition omission.

[0066] Rule 2:

[0067]

[0068] When the quantitative evaluation index values involved meet this rule, it means that the merchant name translation model may have mistakenly identified other fields as address fields, resulting in the final translation result length being too short (since the address field usually does not need to be translated, the length of the direct translation result and the length of the translation result do not include the address field).

[0069] The evaluation result corresponding to the translation result that meets this rule is too many address recognitions.

[0070] Rule 3:

[0071]

[0072] According to the invention patent application with application number 202110426399.X, the merchant name translation model has an industry library. Here, the number of known industries is the number of industries recorded in the industry library in the merchant information.

[0073] When the quantitative evaluation index values involved meet this rule, it means that the merchant name translation model may have mistakenly identified the enterprise name field as the industry field, causing the industry field to be too long.

[0074] The evaluation result corresponding to the translation result that meets this rule is that the industry identification is incorrect.

[0075] Rule 4:

[0076] ((Number of industries ≥ 3 ∩Number of known industries ≤ 1) ∪ (Number of industries ≥ 1 ∩Number of known industries = 0) ∪Number of translated words < 2) ∩Translation retention score < 0.7

[0077] When the quantitative evaluation index values involved meet this rule, it means that the direct translation result of the industry field of the merchant information may be too long. Therefore, the merchant name translation model truncates the industry field too much to meet the length constraint.

[0078] The evaluation result corresponding to the translation result that meets this rule is too much industry truncation.

[0079] Based on the above content, when Rule 1 is met and the evaluation result is address recognition omission, the corresponding optimization direction is to improve the address recognition rules of the merchant name translation model; when Rule 2 is met and the evaluation result is excessive address recognition, the corresponding optimization direction is to improve the address recognition rules of the merchant name translation model; when Rule 3 is met and the evaluation result is incorrect industry identification, the corresponding optimization direction is to improve the industry identification rules of the merchant name translation model; when Rule 4 is met and the evaluation result is excessive industry truncation, the corresponding optimization direction is to improve the truncation rules of the merchant name translation model, or to improve the vocabulary of the merchant name translation model to simplify the translation results corresponding to the industry field.

[0080] It should be made clear here that the technical solution protected by this application is only for determining the optimization direction, that is, by quantitatively evaluating the translation process and translation results of the merchant name translation model according to quantitative evaluation rules, the possible problems in the translation process and translation results are determined, and then the evaluation results and the optimization direction based on the evaluation results are given. The optimization direction only defines a smaller range for the optimization content, and does not determine the specific optimization content of the merchant name translation model, nor does it perform specific optimization on the merchant name translation model.

[0081] Taking the evaluation result of incorrect address recognition as an example, the determined optimization direction is to improve the address recognition algorithm in the merchant name translation model. Then, relevant technical personnel can review the translation results involved one by one or conduct random reviews based on the evaluation result and optimization direction to determine whether the address recognition is indeed incorrect. If the evaluation result is correct, the relevant technical personnel will conduct specific optimization and improvement of the address recognition algorithm in the merchant name translation model.

[0082] Based on the above, it is understandable that the quantitative evaluation rules evaluate quantifiable translation results. For some translation problems, the quantitative evaluation rules may not be able to determine the optimization direction. For example, when the merchant information is very long but the translation result is very short, and there are no obvious abnormalities in the values of other quantitative evaluation indicators, the quantitative evaluation rules cannot provide a clear optimization direction. In this case, the translation result is determined to be an unknown problem and manually determined by relevant technical personnel. However, it should be clear that such position problems account for a very small proportion, and the technical solution protected by this application can determine the corresponding optimization direction for most translation problems.

[0083] The aforementioned quantitative evaluation of the translation process and results utilizes an unsupervised learning approach. By matching quantitative evaluation indicators with relevant rules, merchant information with translation problems and their corresponding translation results are identified, further confirming the difficulties in the translation process. Furthermore, the quantitative evaluation rules categorize translation difficulties, thereby establishing consistent optimization directions for the same type of translation problems and difficulties. This allows for more efficient determination of manual optimization directions. Compared to manual investigation and identification of translation problems and difficulties, this significantly reduces the scope and time cost of investigation, accelerating the optimization process. Furthermore, each optimization direction is determined based on the merchant information to be translated. In other words, the object of unsupervised learning is business data. Therefore, the optimization direction is determined without interrupting external business services.

[0084] In another preferred embodiment, see Figure 2 As shown, Figure 2 Steps S23, S24 and S25 in Figure 1 Steps S12, S13, and S14 in the embodiment are identical or substantially identical, and are therefore omitted for clarity and are incorporated herein by reference. Step S21 includes determining a vocabulary optimization direction for the merchant name translation model based on the merchant information, the vocabulary, and a preset filtering threshold.

[0085] Here, the merchant information translation model includes a vocabulary library, which stores high-frequency fields in the identified merchant information, and their corresponding direct translation results and translation results.

[0086] Optimizing the vocabulary involves determining the correct translation results for commonly used or important fields to be translated and storing them in the vocabulary. This allows you to directly match the corresponding translation results in the vocabulary when translating the field later, reducing machine translation time. Furthermore, for fields with common translation results, you can manually set the translation results to the common translation results to improve translation accuracy and avoid translation errors caused by inconsistencies between the translation results and the common translation.

[0087] Furthermore, based on the merchant information and the vocabulary, a field to be optimized is determined; the frequency of occurrence of the field to be optimized in all merchant information is counted; and based on the frequency of occurrence and a preset filtering threshold, a vocabulary optimization direction of the merchant name translation model is determined.

[0088] Here, the merchant information to be translated is compared with the vocabulary to identify content not stored in the vocabulary. This is temporarily designated as a field to be optimized. The frequency of occurrence of this field in all merchant information to be translated is then counted. Based on the comparison of this frequency with a preset filtering threshold, it is determined whether the field to be optimized is high-frequency content not stored in the vocabulary. If so, the field to be optimized can be determined as content to be expanded in the vocabulary. Specifically, the field to be optimized can be stored in a separate space, which relevant technicians can regularly access to determine whether the field to be optimized stored therein needs to be assigned a corresponding fixed translation result and added to the vocabulary.

[0089] Furthermore, the field to be optimized has a characteristic word, and determining the field to be optimized based on the merchant information and the vocabulary includes: screening out merchant information with the characteristic word from all merchant information; and determining the field in the merchant information that is located before the characteristic word, has irrelevant information deleted, and is not stored in the vocabulary as the field to be optimized, wherein the irrelevant information includes symbols and punctuation.

[0090] Here, because merchant information often contains characteristic words, for example, if the content to be translated is Chinese, when the merchant information represents a business, it will generally contain the word "company." The content before "company" is usually the company name. Therefore, based on the characteristic word "company," the field preceding it can be initially identified as the field to be optimized, and the company name information can be filtered based on this field.

[0091] Still taking the Chinese content to be translated as an example, accordingly, when the vocabulary of the merchant information translation model lacks the enterprise field information of important merchants and the translation results of the enterprise field information need to be optimized in a targeted manner, "company" can be directly set as a feature word to filter out merchant information with this feature word. According to the word order of Chinese, the enterprise name must be located before the feature word "company". Therefore, first compare the field located before the feature word with the vocabulary, retain the fields not stored in the vocabulary, and delete the irrelevant information such as symbols and punctuation therein, and use the remaining information as the field to be optimized. In this way, content suspected of being enterprise name information can be filtered out from the merchant information. Furthermore, the content suspected of being enterprise name information can be stored in a separate storage space, and the relevant technical personnel can determine whether it is enterprise name information that needs to be added to the vocabulary. If there is a universal translation for the enterprise name information, its translation result is manually determined as the universal translation result.

[0092] It should be clear that the technical solution protected by this application does not restrict specific application scenarios. In addition to "company", other feature words that can determine the fields to be optimized are all within the scope of protection of this application, such as "institution", "organization", "research institute", "branch", "campus" and "center". This application generally does not restrict the translation and the language to be translated. When the language to be translated is other than Chinese, the position of the field to be optimized corresponding to a certain feature word must be determined according to the word order of the language.

[0093] This method can be used to determine the optimization direction of the vocabulary in a targeted manner to improve translation speed and accuracy. Taking the predetermined optimization content of corporate name information and Chinese-to-English translation as an example, when translating corporate name information, the merchant information translation model generally uses the pinyin content corresponding to the corporate name as the translation result. However, for well-known companies, they generally have fixed English translations. For example, the English translation of "Lancome" is "lancome." Therefore, when translating such corporate names, directly translating according to the pinyin will result in translation errors. To address this issue, the vocabulary optimization method of this application can be used to screen out frequently appearing corporate names, and relevant technical personnel can determine whether it is necessary to add them to the vocabulary and their corresponding fixed translations.

[0094] Furthermore, because this application uses feature words to filter the fields to be optimized, rather than directly filtering high-frequency fields not stored in the vocabulary, it can prevent individuals or other social organizations from maliciously plagiarizing and creating similar or identical corporate names by filtering with feature words, thereby preventing the maliciously plagiarized content from being translated into the inherent translation of the original company. For example, if a business information contains a corporate name such as "Lancome Hair Salon" that is clearly plagiarized, when filtering with "company" as the feature word, since it does not contain the word "company", its translation will still be "lankou", rather than the inherent "lancome". Therefore, the difference in translation results can distinguish such maliciously plagiarized corporate names from the original corporate names.

[0095] Furthermore, the field to be optimized does not have a feature word, and the determining of the field to be optimized based on the merchant information and the vocabulary includes: segmenting the merchant information according to a word segmentation algorithm to obtain multiple fields; and determining the field in the merchant information that does not have a feature word and is not stored in the vocabulary as the field to be optimized.

[0096] Here, since some fields do not have characteristic words, when it is necessary to optimize such fields in a targeted manner, an exclusion method is adopted to determine the content of the fields without characteristic words. First, the entire merchant information is segmented. The NLP segmentation algorithm can be used for segmentation. Here, no specific segmentation algorithm is specified. After obtaining multiple fields after segmentation, the vocabulary of the merchant information translation model is used to determine the fields that are not stored in the vocabulary, do not have characteristic words, and have an occurrence frequency exceeding the preset filtering threshold. These fields are identified as fields to be optimized.

[0097] In another specific embodiment, the merchant information is first cleaned, punctuation marks and other irrelevant information are deleted, the cleaned information is segmented, and the frequency of occurrence of each field is counted. If it exceeds the preset filtering threshold, it is compared with the vocabulary respectively, and a label is added for each filtered field. Common labels include: whether the length is too short (for example, less than 1), whether it is an address, whether it is a company, whether it is a self-employed person, whether it is in a known industry library, and whether it is in a blacklist (a determined non-industry vocabulary), etc. If an existing label cannot be added, it is marked as unknown. Then the content marked as unknown is used as a field to be optimized that is suspected to be an industry field. Through this exclusion method, the field to be optimized that is suspected to be an industry field is determined and stored in a separate space. The space is accessed by relevant technical personnel to determine whether the field to be optimized stored therein is an industry field, and then the corresponding fixed translation result is set and added to the vocabulary. In this way, the determination of the optimization direction for industry information is achieved.

[0098] In addition, still taking Chinese-English translation as an example, the English literal translation results corresponding to industry information are generally long. For example, the literal translation result corresponding to "mother and baby" can be "Maternal and infant". If the merchant information translation model directly uses this as the translation result, under the condition of limiting the length of the translation result, it will cause the industry information to be too long and thus compress other valid information, resulting in a poor translation result with too little valid information. Therefore, through the solution of this application, the industry can be screened out, and then the relevant technical personnel can determine the corresponding translation result, for example, "Maternal and infant" can be abbreviated to "infant&Mom", etc. Thus, while complying with the length limit of the translation result, the effectiveness of the translation result is improved and invalid translation is avoided.

[0099] In addition, the filtering threshold can be determined by demand. Specifically, when it is necessary to expand the vocabulary on a large scale, the filtering threshold can be appropriately lowered to add the medium and high frequency fields in the merchant information to the content to be optimized. Accordingly, the filtering threshold is lower, the number of fields to be optimized is higher, and the optimization workload is also greater; and when the server's computing volume is smaller, or only the vocabulary of important fields needs to be expanded according to demand, the filtering threshold can be appropriately increased, and only the ultra-high frequency fields in the merchant information are added to the content to be optimized. Correspondingly, the filtering threshold is higher, the number of fields to be optimized is smaller, and the optimization workload is also less.

[0100] Furthermore, step S22 includes: obtaining translation results of the merchant name translation model optimized according to the vocabulary optimization direction for multiple merchant information.

[0101] In this step, after optimizing the business information translation model using the aforementioned lexicon optimization direction, the business information translation results are determined based on the optimized business information translation model, and then the translation results are quantitatively evaluated as described above. This method, based on the lexicon-optimized business information translation model, quantitatively evaluates the translation results and determines further optimization directions. This not only improves the quality of the translation results, but also accelerates the optimization process, allowing the business information translation model to more quickly approach the ideal model.

[0102] Compared to existing technologies, this application obtains the translation results of the merchant name translation model for multiple merchant information; determines the quantitative evaluation index values for each translation result based on multiple preset quantitative evaluation indicators; determines the evaluation result corresponding to each translation result using preset quantitative evaluation rules based on the quantitative evaluation index values; and determines the optimization direction of the merchant name translation model corresponding to the evaluation result. This method determines the optimization direction of the merchant name translation model, allowing relevant technical personnel to determine specific optimization content based on this optimization direction, significantly reducing the time and workload required for manual investigation and saving the optimization cost of the merchant name translation model.

[0103] In addition, an embodiment of the present application further provides a computer-readable medium on which computer-readable instructions are stored. The computer-readable instructions can be executed by a processor to implement the aforementioned method.

[0104] The embodiment of the present application further provides a device for determining an optimization direction of a merchant name translation model, wherein the device includes:

[0105] one or more processors; and

[0106] A memory storing computer-readable instructions that, when executed, cause the processor to perform the operations of the aforementioned method.

[0107] For example, when the computer-readable instructions are executed, the one or more processors may: obtain translation results of the merchant name translation model for a plurality of merchant information;

[0108] Determining the quantitative evaluation index values of each translation result based on a plurality of preset quantitative evaluation indexes;

[0109] Determining the evaluation result corresponding to each translation result according to the quantitative evaluation index value using a preset quantitative evaluation rule;

[0110] Determine an optimization direction of the merchant name translation model corresponding to the evaluation result.

[0111] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above and that the invention can be implemented in other specific forms without departing from the spirit or essential features of the invention. Therefore, from any point of view, the embodiments should be regarded as exemplary and non-restrictive, and the scope of the invention is defined by the appended claims rather than the above description, and it is intended that all changes that fall within the meaning and scope of the equivalents of the claims be encompassed within the present invention. Any figure marks in the claims should not be regarded as limiting the claims involved. In addition, it is clear that the word "comprising" does not exclude other units or steps, and the singular does not exclude the plural. Multiple units or devices stated in the device claim may also be implemented by one unit or device through software or hardware. Words such as first and second are used to indicate names and do not indicate any particular order.

Claims

1. A method for determining an optimization direction of a merchant name translation model, wherein: The method comprises: Obtain translation results of the merchant name translation model for multiple merchant information; Determining the quantitative evaluation index values of each translation result based on a plurality of preset quantitative evaluation indexes; Determining an evaluation result corresponding to each translation result according to the quantitative evaluation index value using a preset quantitative evaluation rule, the quantitative evaluation index including at least one of the following: address length, company name length, industry length, number of industries, merchant information length, length of a literal translation result not truncated according to a translation length limit, length of a translation result truncated according to a translation length limit, number of translated words, and a translation retention score, wherein the translation retention score is a ratio of the translation result length to the literal translation result length, and the evaluation result including at least one of the following: omission of address recognition, excessive address recognition, incorrect industry recognition, and excessive truncation of industries; Determining an optimization direction for the merchant name translation model corresponding to the evaluation result, where the optimization direction corresponding to the evaluation result includes at least one of the following: improving an address recognition rule, an industry recognition rule, a truncation rule, or improving a vocabulary of the merchant name translation model; The quantitative evaluation rules include the following: Rule 1: ∩ Translation retention score < 0.6; Rule 2: ∩The translation result length is less than 10; Rule 3: ∩ Translation retention score < 0.6; Rule 4: ((number of industries ≥ 3 ∩ number of known industries ≤ 1) ∪ (number of industries ≥ 1 ∩ number of known industries = 0) ∪ number of translated words < 2) ∩ translation retention score < 0.7; When Rule 1 is met, the evaluation result is omission of address recognition, and the corresponding optimization direction is to improve the address recognition rules of the merchant name translation model; when Rule 2 is met, the evaluation result is excessive address recognition, and the corresponding optimization direction is to improve the address recognition rules of the merchant name translation model; when Rule 3 is met, the evaluation result is incorrect industry identification, and the corresponding optimization direction is to improve the industry identification rules of the merchant name translation model; when Rule 4 is met, the evaluation result is excessive industry truncation, and the corresponding optimization direction is to improve the truncation rules of the merchant name translation model, or to improve the vocabulary of the merchant name translation model to simplify the translation results corresponding to the industry field.

2. The method according to claim 1, wherein The merchant name translation model includes a vocabulary, and before obtaining the translation results of the merchant name translation model for the plurality of merchant information, the method further includes: Determining a vocabulary optimization direction of the merchant name translation model based on the merchant information, the vocabulary, and a preset filtering threshold; The step of obtaining translation results of the merchant name translation model for the plurality of merchant information includes: Obtain translation results of the merchant name translation model optimized according to the vocabulary optimization direction for multiple merchant information.

3. The method according to claim 2, wherein: The determining of the vocabulary optimization direction of the merchant name translation model based on the merchant information, the vocabulary, and a preset filtering threshold includes: Determining a field to be optimized based on the merchant information and the vocabulary; Count the occurrence frequency of the field to be optimized in all merchant information; Based on the occurrence frequency and a preset filtering threshold, a vocabulary optimization direction of the merchant name translation model is determined.

4. The method according to claim 3, wherein: The field to be optimized has a characteristic word, and determining the field to be optimized based on the merchant information and the word library includes: Filtering out merchant information having the characteristic words from all merchant information; The field in the merchant information that is located before the feature word, has irrelevant information deleted, and is not stored in the vocabulary is determined as the field to be optimized, wherein the irrelevant information includes symbols and punctuation marks.

5. The method according to claim 3, wherein The field to be optimized does not have a characteristic word, and determining the field to be optimized based on the merchant information and the vocabulary includes: Segmenting the merchant information according to a segmentation algorithm to obtain multiple fields; The field in the merchant information that does not have a feature word and is not stored in the word library is determined as the field to be optimized.

6. The method according to any one of claims 3 to 5, wherein determining the vocabulary optimization direction of the merchant name translation model based on the occurrence frequency and a preset filtering threshold comprises: The fields to be optimized whose occurrence frequency exceeds a preset filtering threshold are used as the content to be expanded in the vocabulary. 7 . A computer-readable medium having computer-readable instructions stored thereon, wherein the computer-readable instructions can be executed by a processor to implement the method according to claim 1 .

8. A device for determining a translation model optimization direction, wherein: The device includes: one or more processors; and A memory storing computer readable instructions which, when executed, cause the processor to perform the operations of the method of any one of claims 1 to 6.

Images