Object rating method and device, equipment and storage medium

Abstract

The invention discloses an object rating method, device and equipment and a storage medium, and the method comprises the steps: constructing a rating method set based on an existing rating report, and determining a target rating item of a to-be-rated object according to the rating method set; determining a first score of a first target score item of a first quantization category in the target score items based on a language model; determining a second score of a second target score item of a second quantization category in the target score items through a quantization method; and according to the first score and the second score, determining a score level of the to-be-rated object. According to the method, the rating method set is taken as priori knowledge, the score of the non-quantizable score item is determined through the large model, and the score of the quantizable score item is calculated through the quantized data in the rating method set, so that the labor cost and the time cost required by manual rating can be reduced, and the rating efficiency is improved. And the accuracy of the enterprise object rating and the rating consistency with the rated enterprise object are also improved.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence technology, and in particular to an object rating method, device, equipment and storage medium. Background Art

[0002] With the continuous development of financial technology, especially Internet technology finance, more and more technologies are applied in the financial field, but the financial industry has also put forward higher requirements for technology, such as the financial industry has higher requirements for corporate ratings. However, in the existing technology, financial practitioners are required to rate companies. This method not only has the problem of strong subjectivity, but also requires raters to spend a lot of time to screen and analyze resources related to corporate operations, which affects the efficiency and accuracy of corporate ratings.

[0003] Therefore the prior art still needs to be improved and enhanced. Summary of the invention

[0004] The technical problem to be solved by the present application is to provide an object rating method, device, equipment and storage medium in view of the deficiencies in the prior art.

[0005] In order to solve the above technical problems, the first aspect of the present application provides an object rating method, wherein the object rating method specifically includes: Constructing a rating method set based on an existing rating report, wherein the rating method set includes a plurality of rating methods and a plurality of scoring items corresponding to each rating method, wherein the scoring items include a plurality of rating data target rating methods; Determine the target rating method corresponding to the to-be-rated object and the target scoring item corresponding to the target rating method according to the rating method set; Acquire a first target scoring item of a first quantization category in the target scoring items, determine first target scoring data for the first target scoring item, and score the to-be-rated object based on the first target scoring data and a preset large language model to determine a first score; Acquire a second target scoring item of a second quantitative category in the target scoring item, determine second target scoring data for the second target scoring item, and score the to-be-rated object based on the second target scoring data to determine a second score; The rating level of the object to be rated is determined according to the first rating and the second rating.

[0006] The object rating method, wherein the rating method set constructed based on the existing rating report specifically includes: Obtain several existing rating reports; Respectively obtain the detection target of each existing rating report and the detection data corresponding to the detection target, wherein the detection target includes a title, a paragraph and a table, and the detection data includes category information, location information and text content; Determine, based on the detection data of all detection targets of each existing rating report, a rating method adopted by each existing rating report, a number of rating items corresponding to the rating method, and a plurality of rating data of the rating items; A rating method set is constructed according to the determined rating methods, a number of rating items corresponding to each rating method, and a plurality of rating data of the rating items.

[0007] The object rating method, wherein the step of respectively acquiring the detection target of each existing rating report and the detection data corresponding to the detection target, specifically includes: Input each of the existing rating reports into a trained text layout detection model, and output the detection targets in each of the existing rating reports and the category information and location information of the detection targets through the text layout detection model; Extracting text content of each of the detection targets from the corresponding existing rating reports based on the location information of each of the detection targets; The category information, the location information and the text content are used as detection data of the detection target, so as to obtain the detection target of each existing rating report and the detection data corresponding to the detection target.

[0008] The object rating method, wherein the trained text layout detection model specifically includes: Acquire a text layout dataset, wherein the text layout dataset includes a plurality of training texts and annotation target information corresponding to the training texts; The preset target detection model is trained using the training text in the text layout data to obtain a trained text layout detection model.

[0009] The object rating method, wherein the step of determining the rating method adopted by each existing rating report, the several rating items corresponding to the rating method, and the multiple rating data of the rating items based on the detection data of all detection targets of each existing rating report specifically includes: Determining the rating method adopted by the existing rating report based on the detection data of the natural segment; Based on the detection data of the detection target, several scoring items and multiple scoring data of the scoring items corresponding to the rating method are determined.

[0010] The object rating method, wherein the rating data includes a rating item name, a rating value, a rating basis and quantitative data, and the detection data based on the detection target determines a number of rating items and the rating data of the rating items corresponding to the rating method, specifically including: Determine a number of scoring items corresponding to the rating method and the scoring item name and scoring value of each scoring item based on the detection data in the table; Determine the scoring basis of each scoring item based on the detection data of the title and the natural paragraph; The rating quantification method corresponding to each rating item is obtained, and the quantitative data of each rating item is determined through a preset large language model based on the rating quantification method.

[0011] In the object rating method, the step of determining the scoring basis of each scoring item based on the detection data of the title and the paragraph specifically includes: Determine the target title corresponding to each rating item, and obtain the target paragraph corresponding to the target title; Based on the text content in the detection data of the target title and the text content in the detection data of the target paragraph, a scoring basis for the scoring item name is determined.

[0012] In the object rating method, the step of determining the target title corresponding to each rating item and obtaining the target paragraph corresponding to the target title specifically includes: Obtaining a structured directory of the existing rating report, wherein the structured directory consists of multiple levels of titles and contents contained in each level of titles; Determine the target title corresponding to each scoring item according to the scoring item name of each scoring item and the structured directory; The target paragraph corresponding to the target title is determined according to the report content contained in the target title.

[0013] The object rating method, wherein the process of constructing the structured directory specifically includes: Determine the title level of each title in the existing rating report, and determine the subordinate relationship between the titles and the content contained in each title based on the title level of each title and the position information of each title; Based on the subordinate relationships between the titles and the content contained in each title, a structured directory of the existing rating reports is constructed.

[0014] In the object rating method, the step of determining the title level of each title in the existing rating report specifically includes: Get the title font size and title sequence number of each title; Based on the obtained font size and title sequence number of each title, the title level of each title is determined.

[0015] In the object rating method, the process of determining the content contained in each title specifically includes: Based on the position information of each detection target, the detection targets are sorted according to the order in the existing rating report to obtain a detection target sequence; The detection target and detection data between two adjacent titles of the same level in the detection target sequence are taken as the contents contained in the title sorted first.

[0016] The object rating method, wherein the rating of the object to be rated based on the first target rating data and a preset large language model, and determining the first rating specifically includes: Obtaining several keywords corresponding to the first target rating item; Determining summary texts about the keywords based on the object name of the object to be rated and the keywords; Based on the summary text, selecting first target scoring data for the first target scoring item in the rating method set; Based on the summary text and the first target rating data, the to-be-rated object is rated by using a preset large language model to determine a first rating.

[0017] The object rating method, wherein the rating of the object to be rated is performed based on the summary text and the first target rating data by using a preset large language model, and determining the first rating specifically includes: Constructing a scoring prompt word for the first target scoring item; The scoring prompt word, the first target scoring item, the first target scoring data of the first target scoring item and the summary text are spliced ​​and input into a preset large language model, and the first score of the first target scoring item is output through the preset large language model.

[0018] In the object rating method, the step of obtaining a number of keywords corresponding to the first target rating item specifically includes: Selecting all the scoring data corresponding to the first target scoring item in the rating method set, and segmenting the scoring basis in the first target scoring data to obtain a plurality of words; Calculate the similarity between each word and the scoring item name of the first target scoring item and the word frequency-inverse document frequency of each word respectively; Determining the importance of each word based on the similarity of each word and the word frequency-inverse document frequency; A number of keywords are selected from the number of words based on the importance of each word to obtain a number of keywords corresponding to the first target scoring item.

[0019] In the object rating method, the step of determining summary texts about the keywords based on the object name of the object to be rated and the keywords specifically includes: Searching for text resources of the object to be rated based on the object name and a number of keywords of the object to be rated; Dividing the searched text resources into a number of natural segments, and selecting a preset number of representative natural segments from the natural segments based on the object name and a number of keywords; Based on the preset number of representative natural paragraphs and the several keywords, summary texts about the several keywords are determined by using a preset large language model.

[0020] The object rating method, wherein rating the object to be rated based on the second target rating data, and determining the second rating specifically comprises: Acquire quantitative data in the second target scoring data, wherein the quantitative data includes a code function corresponding to a rating quantification method and a quantitative index required by the rating quantification method; Obtaining quantitative indicator data corresponding to the quantitative indicator; The quantitative indicator data is used as an input of the code function to obtain a second score for the second target scoring item.

[0021] The object rating method, wherein determining the rating level of the object to be rated according to the first rating and the second rating specifically comprises: Weighting the first score and the second score to obtain a final score; According to the preset correspondence between the rating level and the rating, the rating level corresponding to the final rating is determined to obtain the rating level of the object to be rated.

[0022] A second aspect of the present application provides an object rating device, wherein the object rating device specifically includes: A construction module, used to construct a rating method set based on an existing rating report, wherein the rating method set includes a plurality of rating methods and a plurality of scoring items corresponding to each rating method, wherein the scoring items include a plurality of rating data target rating methods; A rating method determination module, used to determine a target rating method corresponding to the to-be-rated object and a target scoring item corresponding to the target rating method according to the rating method set; A first processing module is used to obtain a first target scoring item of a first quantization category in the target scoring item, determine first target scoring data for the first target scoring item, and score the to-be-rated object based on the first target scoring data and a preset large language model to determine a first score; A second processing module is used to obtain a second target scoring item of a second quantitative category in the target scoring item, determine second target scoring data for the second target scoring item, and score the to-be-rated object based on the second target scoring data to determine a second score; A grade determination module is used to determine the grade of the object to be rated according to the first score and the second score.

[0023] A third aspect of the present application provides a computer-readable storage medium, which stores one or more programs. The one or more programs can be executed by one or more processors to implement the steps in any of the object rating methods described above.

[0024] A fourth aspect of the present application provides a terminal device, comprising: a processor and a memory; The memory stores a computer-readable program executable by the processor; When the processor executes the computer-readable program, the processor implements the steps in any of the object rating methods described above.

[0025] Beneficial effects: 1. This application uses the object rating method extracted from the evaluation report of existing objects (such as enterprises) as prior knowledge, and determines the rating level of the object to be rated (such as an enterprise) through a large language model, reducing the dependence of object rating on the professional ability of the rater.

[0026] 2. This application screens and integrates the relevant information of the rated object (such as an enterprise) based on the scoring items to determine the key resource information used for rating. This can reduce the time spent on screening and analyzing resources related to the object's operation, and improve the efficiency and accuracy of rating the enterprise object.

[0027] 3. This application uses a summary text reflecting the business conditions to select target scoring data for objects with the same business status as the object to be rated from all scoring data, and uses these target scoring data as prior knowledge of the large model, which can ensure the consistency of scoring results for objects with the same business conditions and improve the rating accuracy of the object to be rated. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0029] Figure 1A flowchart of an object rating method provided in an embodiment of the present application.

[0030] Figure 2 A flowchart of the process of constructing a rating method set.

[0031] Figure 3 FIG. 4 is a flow chart of an implementation method of step S30 .

[0032] Figure 4 FIG. 4 is a flow chart of an implementation method of step S40 .

[0033] Figure 5 This is a functional block diagram of an object rating device provided in an embodiment of the present application.

[0034] Figure 6 A functional block diagram of a terminal device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0035] The embodiments of the present application provide an object rating method, device, equipment and storage medium. In order to make the purpose, technical solution and effect of the present application clearer and more specific, the present application is further described in detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described here are only used to explain the present application and are not used to limit the present application.

[0036] It will be understood by those skilled in the art that, unless expressly stated, the singular forms "one", "said", and "the" used herein may also include plural forms. It should be further understood that the term "comprising" used in the specification of the present application refers to the presence of the features, integers, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or groups thereof. It should be understood that when we refer to an element as being "connected" or "coupled" to another element, it may be directly connected or coupled to the other element, or there may be an intermediate element. In addition, the "connection" or "coupling" used herein may include wireless connection or wireless coupling. The term "and / or" used herein includes all or any unit and all combinations of one or more associated listed items.

[0037] It will be understood by those skilled in the art that, unless otherwise defined, all terms (including technical and scientific terms) used herein have the same meaning as those generally understood by those skilled in the art to which this application belongs. It should also be understood that terms such as those defined in general dictionaries should be understood to have meanings consistent with those in the context of the prior art, and will not be interpreted with idealized or overly formal meanings unless specifically defined as here.

[0038] It should be understood that the sequence numbers and sizes of the steps in this embodiment do not mean the order of execution. The execution order of each process is determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiment of the present application.

[0039] The application content is further explained below through the description of embodiments in conjunction with the accompanying drawings.

[0040] This embodiment provides an object rating method, such as Figure 1 As shown, the method includes: S10. Construct a rating methodology set based on existing rating reports.

[0041] Specifically, the rating method set is used as a priori knowledge database of the object rating method, and the target rating method corresponding to the rating of the object to be rated can be selected from the rating method set. Among them, the rating method set includes several rating methods, each rating method includes several rating items, each rating item includes multiple rating data, and the several rating items included in the rating method are different, and different rating methods can include the same rating items. For example, the rating method set includes rating method A and rating method B, rating method A includes rating item A1, rating item A2 and rating item A3, rating method B includes rating item B1, rating item B2 and rating item B3, rating item A1, rating item A2 and rating item A3 are different from each other, rating item B1, rating item B2 and rating item B3 are different from each other, and rating item A1 and rating item B2 are the same.

[0042] Furthermore, the multiple scoring data may include a scoring item name, a scoring value, a scoring basis, and quantitative data, wherein the scoring item name is a unique identifier of the scoring item, the scoring value is a rating score of the scoring item, the scoring basis is a textual explanation of the scoring item to obtain its corresponding rating score, and the quantitative data includes the quantitative category of the scoring item, and when the quantitative category is the first quantitative category, the quantitative data also includes the code function of the quantitative method corresponding to the scoring item and the required quantitative indicators. In other words, each evaluation and rating method can rate an object based on multiple rating factor scoring items (i.e., the scoring item name), and each scoring item will have a specific rating score (i.e., the rating value), a textual explanation of the rating score (i.e., the scoring basis), and the quantitative category of the scoring item and the corresponding quantitative method (i.e., the quantitative data).

[0043] In a specific implementation, Figure 2 As shown, the rating method set constructed based on the existing rating report specifically includes: S11. Obtain several existing rating reports; S12, respectively obtaining the detection target of each existing rating report and the detection data corresponding to the detection target, wherein the detection target includes a title, a paragraph and a table, and the detection data includes category information, location information and text content; S13, determining the rating method adopted by each existing rating report, a number of rating items corresponding to the rating method, and a plurality of rating data of the rating items based on the detection data of all detection targets of each existing rating report; S14: construct a rating method set according to the determined rating methods, the several rating items corresponding to the rating methods, and the multiple rating data of the rating items.

[0044] Specifically, in step S11, the existing rating report is a report formed by rating an existing object, for example, a report formed by evaluating and rating an enterprise object. The rated objects corresponding to the existing rating reports may be different, and some of the rated objects may belong to the same industry. At the same time, some of the existing rating reports may use the same rating method, and some of the existing rating reports may use different rating methods. The existing rating reports may be in PDF format or in Word format.

[0045] In addition, after obtaining several existing rating reports, the industry to which the object corresponding to each existing rating report belongs can be obtained, and the industry can be associated with the rating method adopted by the existing rating report to configure the industry category for the evaluation rating method, so as to facilitate the subsequent search for the target rating method corresponding to the object to be evaluated in the rating method set based on the existing rating report. Of course, in actual applications, other category labels can also be configured for the rating method adopted by the existing rating report. For example, the operating status category can be configured for the rating method adopted by the existing rating report according to the operating status of the object corresponding to the existing rating report, or the industry category and operating status category can be configured for the object corresponding to the existing rating report, and then when searching for the target rating method corresponding to the object to be evaluated, the search can be performed according to the industry category first, and then the search results can be searched according to the operating status category, etc.

[0046] Further, in step S12, the detection target includes a title, a paragraph and a table, and the detection data includes category information, position information and text content. The category information is used to reflect that the detection target is a title, a paragraph or a table, the position information is used to reflect the text position of the detection target in the existing rating report, and the text content is used to reflect the text content included in the detection target. Specifically, the position information may include the upper left corner coordinates and the lower right corner coordinates of the detection target, and since the existing rating report may include multiple pages of text, the position information may also include the page number of the text page where the detection target is located. Of course, the position coordinates may also be represented in other ways, for example, the position coordinates may be the center point coordinates, the width of the detection target and the height of the detection target. The text content is the text content included in the detection target, that is, the text content may include the text content. In addition, since in actual applications, the title and the body generally use different font sizes, the font size may also be included in the text content to facilitate the construction of the subsequent directory structure.

[0047] In one implementation, respectively acquiring the detection target of each existing rating report and the detection data corresponding to the detection target specifically includes: S121, inputting each of the existing rating reports into a trained text layout detection model, and outputting the detection targets in each of the existing rating reports and the category information and location information of the detection targets through the text layout detection model; S122, extracting text content of each of the detection targets from the corresponding existing rating reports based on the position information of each of the detection targets; S123: Use the category information, the location information, and the text content as detection data of a detection target, so as to obtain the detection target of each existing rating report and the detection data corresponding to the detection target.

[0048] Specifically, the text layout detection model is a trained neural network model, and each detection target in the existing rating report and the location information of each detection target can be detected through the text layout detection model. That is to say, the input item of the text layout detection model is the existing rating report, and the output item is the detection target information. The detection target information includes the category information of the detection target and the location information of the detection target. The category information includes titles, paragraphs and tables, and the location information includes page numbers, upper left corner coordinates and lower right corner coordinates. That is, the detection target information can be expressed as (page_num, class, x1, y1, x2, y2), where page_num represents the page number, class represents the category information of the detection target, (x1, y1) represents the upper left corner coordinates of the detection target, and (x2, y2) represents the lower right corner coordinates of the detection target.

[0049] Exemplarily, the training process of the text layout detection model may be: Get the text layout dataset; The preset target detection model is trained using the training text in the text layout data to obtain a trained text layout detection model.

[0050] Specifically, the text layout dataset includes several training texts and the annotation target information corresponding to the training texts, the annotation target information includes annotation category information and annotation position information, and each training text includes one or more of a title, a paragraph, and a table. For example, the training text is a single-page financial PDF text, and the annotation target information is (class, x1, y1, x2, y2), where class represents the annotation category information of the detection target (title, table, or paragraph), (x1, y1) represents the coordinates of the upper left corner of the detection target, (x2, y2) represents the coordinates of the lower right corner of the detection target, the coordinate origin is the upper left corner of the single-page financial PDF text, and the x and y axes are the short side and long side of the single-page financial PDF text, respectively.

[0051] Furthermore, when the preset target detection model is trained using the training text in the text layout data, the training text is input into the preset target detection model, and the preset target detection model outputs the predicted target information corresponding to the training text, and the predicted target information includes predicted category information and predicted position information. Then, a loss term (e.g., contrast loss term, L1 loss term, etc.) is constructed based on the predicted target information and the labeled target information, and the preset target detection model is trained based on the loss term to obtain a trained text layout detection model. In addition, in practical applications, the preset target detection model can adopt a pre-trained open source target detection model, such as the YOLO series, etc., and then the pre-trained target detection model is fine-tuned using the text layout dataset.

[0052] It should be noted that when performing target detection on an existing rating report through a trained text layout detection model, the existing rating report can be first split into pages, and the page number of each text page obtained by the split in the existing rating report can be recorded. Then, target detection is performed on each text page through the text layout detection model to obtain the category information and detection position information of the detection target in each text page. Then, the page number of each text page is added to the detection position information to obtain the position information of the detection target.

[0053] After detecting the location information of the detection target, the target document corresponding to the detection target can be determined based on the location information, and then the text content and font size can be obtained by parsing the target document (for example, text recognition, etc.), and then the text content and font size are used as the text content of the detection target. Of course, in practical applications, since existing rating reports generally use PDF format, after obtaining the target document corresponding to the detection target, the PDF parsing tool can be directly used to parse the target document to obtain the text content and font size corresponding to the detection target. Finally, the category information, location information and content information of the detection target are used as the detection data of the detection target, and the detection target of each existing rating report and the detection data corresponding to the detection target are obtained.

[0054] Further, in step S13, after obtaining the detection data of the detection target included in the existing rating report, the rating method used in the existing rating report can be searched in the detection data of the detection target included in the existing rating report, and a number of scoring items corresponding to each evaluation rating method and multiple scoring data for each rating item can be determined.

[0055] Exemplarily, the determining of the rating method adopted by each existing rating report, the several scoring items corresponding to the rating method, and the multiple scoring data of the scoring items based on the detection data of all detection targets of each existing rating report specifically includes: S131, determining the rating method adopted by the existing rating report based on the detection data of the natural segment; S132. Based on the detection data of the detection target, determine a number of scoring items and a plurality of scoring data of the scoring items corresponding to the rating method.

[0056] In step S131, the rating method is generally recorded in the existing rating report in text form, so when determining the rating method used in the existing rating report, the rating method used in the existing rating report will be searched in the detection data of the natural segment. That is to say, the rating method can be used as a search keyword to search for the rating method used in the existing rating report in the text content corresponding to the natural segment. In addition, in actual applications, the rating method can have different versions, so when determining the rating method, the version number of the rating method can also be searched synchronously to more accurately determine the rating method used in the existing rating report. Correspondingly, when determining the search keyword, the rating method and the version number can be used as the search keyword, and then the rating method and its version number used in the existing rating report can be searched in the text content corresponding to the natural segment according to the search keyword, and the rating method and its version number found are used as the rating method of the existing rating report. Of course, in actual applications, the rating method and its version number used in the existing rating report can also be directly searched in the text content of all detection targets according to the search keyword.

[0057] In step S132, after obtaining the rating method, several scoring items corresponding to the rating method and multiple scoring data of each scoring item are searched in the detection data of all detection targets corresponding to the existing rating reports. For example, the scoring item names of the scoring items included in the rating method can be determined first, and then the scoring values ​​of each scoring item can be determined based on the scoring item names, and then the scoring basis and quantitative data of each scoring item can be determined.

[0058] Based on this, the detection data based on the detection target determines a number of scoring items and scoring data of the scoring items corresponding to the rating method, specifically including: S1321, determining a number of scoring items corresponding to the rating method and a scoring item name and a scoring value of each scoring item based on the detection data in the table; S1322, determining a scoring basis for each scoring item based on the detection data of the title and the natural paragraph; S1323: Obtain a rating quantization method corresponding to each rating item, and determine the quantitative data of each rating item through a preset large language model based on the rating quantization method.

[0059] In step S1321, in existing rating reports, rating factors and scores are generally presented in a table format, so that the rating item names included in the rating method and the rating values ​​corresponding to each rating item name can be searched in the text content of the table. Specifically, the rating factors and scores can be used as search keywords to search for the rating item names and rating values ​​of each rating item included in the rating method in the text content of the table, wherein the column where the "rating factor" is located is the rating item name of the rating method, each rating item name corresponds to a score, and the column where the "score" is located is the rating value corresponding to the rating item.

[0060] S1322. After obtaining the name and score of the scoring item, in order to enable the large model to better understand the reason for the scoring value for the scoring item, the scoring basis of each scoring item can be obtained, and a textual explanation of the scoring reason can be given through the scoring basis. The scoring basis can be obtained by searching the text content of the detection target. In actual applications, the scoring basis will be recorded in a natural paragraph, so when obtaining the scoring basis, the scoring basis can be determined by the text content of the natural paragraph used to explain the scoring item.

[0061] Exemplarily, the scoring basis for determining each scoring item based on the detection data of the title and the paragraph specifically includes: Determine the target title corresponding to each rating item, and obtain the target paragraph corresponding to the target title; Based on the text content in the detection data of the target title and the text content in the detection data of the target paragraph, a scoring basis for the scoring item name is determined.

[0062] Specifically, the target title is a title in an existing rating report, and the target title includes the rating item name corresponding to the rating item. That is to say, the rating item name can be used as a keyword to search for the target title corresponding to each rating item in all the titles in the existing rating report, and then the target natural segment corresponding to the target title is obtained, wherein the target natural segment is a natural segment belonging to the target title.

[0063] Exemplarily, determining the target title corresponding to each rating item and obtaining the target paragraph corresponding to the target title specifically includes: Obtaining a structured directory of the existing rating reports; Determine the target title corresponding to each scoring item according to the scoring item name of each scoring item and the structured directory; The target paragraph corresponding to the target title is determined according to the report content contained in the target title.

[0064] Specifically, the structured directory is composed of multiple levels of titles and the contents contained in each level of titles, wherein the structured directory structure can adopt a tree structure in json format.

[0065] For example, the directory structure can be: [ "First level title name":{ "Natural paragraph content":..., "Table Contents": ..., "Subtitle": [ "Secondary Title Name":{ "Natural paragraph content":..., "Table Content":..., "Subtitle":[ "Level 3 Title Name":{ "Natural paragraph content": ..., "Table Contents": ..., }, ] }, ] }, ] Exemplarily, the process of constructing the structured directory specifically includes: N10. Determine the title levels of each title in the existing rating report, and determine the subordination relationship between titles and the content included in each title based on the title levels of each title and the position information of each title; N20. Based on the subordination relationship between titles and the content included in each title, construct a structured table of contents for the existing rating report.

[0066] In step N10, the title level is obtained by grading the title and is used to reflect which level of title it is. For example, the title levels include first-level titles, second-level titles, third-level titles, etc. Among them, the title level can be determined according to the font size and serial number in the text content of the title.

[0067] Exemplarily, the specific steps of determining the title levels of each title in the existing rating report include: Obtain the title font size and title serial number of each title; Based on the obtained title font size and title serial number of each title, determine the title level of each title.

[0068] Specifically, the title font size and title serial number can be obtained by reading the text content of the title. The title serial number is the first text in the text content of the title, such as "(1)", "1.", etc. The title font size is the font size used for the title. After obtaining the title font size and title serial number, the title levels of each title can be determined according to the preset title level conditions. Among them, the title level conditions can be: if the title font size is greater than the average font size of the natural paragraph and has no serial number, then the title level of this title is the first level; if the title font size is equal to the average font size of the natural paragraph and the title serial number is the serial number in the first preset form (such as the serial number in the form of "(1)"), then the title level of this title is the second level; if the title font size is equal to the average font size of the natural paragraph and the title serial number is the serial number in the first preset form (such as the serial number in the form of "1."), then the title level of this title is the third level.

[0069] It should be noted that in practical applications, other methods can also be used to determine the title level. For example, input the text content of the title into a large model and output the title level through the large model.

[0070] Furthermore, the subordination relationship between titles refers to that between two adjacent titles with the same title level (denoted as the first title), and the title with a title level one level lower than the first title (denoted as the second title) is the sub-title of the first title in front of the two first titles. For example, the second-level title between two adjacent first-level titles is the sub-title of the first-level title in front, and the third-level title between two adjacent second-level titles is the sub-title of the second-level title in front.

[0071] The content included in each title includes the text content included in the title itself, the text content included in the subtitles that have a subordinate relationship with each title, and the natural paragraph content located after each title and its subtitles. In other words, the content included in each title includes the text content of all detection targets between the title and the specific title and the text content included in the title itself, wherein the specific title is an adjacent title with the same title level as the title and located after the title.

[0072] Exemplarily, the process of determining the content contained in each title specifically includes: Based on the position information of each detection target, the detection targets are sorted according to the order in the existing rating report to obtain a detection target sequence; The detection target and detection data between two adjacent titles of the same level in the detection target sequence are taken as the contents contained in the title sorted first.

[0073] Specifically, the detection target sequence includes all detection targets, and each detection target is arranged in the order in which it appears in the existing rating report, that is, for detection target a and detection target b in the detection target sequence, when detection target a is located before detection target b, it means that detection target a appears in the existing rating report before detection target b. The detection target sequence can be obtained by sorting the detection targets with page number as the first keyword and the y-direction coordinate in the upper left corner coordinate as the second keyword, or with page number as the first keyword and the y-direction coordinate in the lower right corner coordinate as the second keyword, etc. In addition, after obtaining the order of each detection target in the existing rating report, the detection targets between two adjacent titles of the same title level are attributed to the title content of the previous title to obtain the content contained in each title, wherein the title content can include natural paragraphs and / or tables.

[0074] Further, in step N20, after obtaining the subordinate relationship between each title and the content contained in each title, each title is established into a title structure according to the subordinate relationship, and then the content contained in each title is added to the back of each title to obtain a structured directory. The above completes the description of the process of constructing a structured directory.

[0075] Further, after obtaining the structured directory, the target titles corresponding to the rating items corresponding to the rating item names are searched in the structured directory using the rating item names as keywords, that is, the title of the highest title level containing the rating item target is searched in the structured directory, and the found title is used as the target title corresponding to the rating item. After obtaining the target title, the content contained in the title is read, and then the natural paragraph in the content contained in the title is obtained to obtain the target natural paragraph corresponding to the target title.

[0076] Furthermore, after obtaining the target title and the target paragraph, the text content in the detection data of the target title and the text content in the detection data of the target paragraph are used as the scoring basis for the scoring item name. Specifically, the text content of the target paragraph is spliced ​​with the text content of the target title, and the spliced ​​text content is used as the scoring basis for the scoring item, wherein the scoring basis may include the business conditions of the enterprise object and other contents. In addition, it should be noted that when the target title includes a subtitle, the paragraph included in the subtitle is also the target paragraph corresponding to the target title.

[0077] Further, in step S1323, after obtaining the scoring item name, scoring value and scoring basis of the scoring item, it can also be determined whether the scoring item is quantifiable. When it is quantifiable, the rating quantification method of the scoring item can also be obtained, and the rating quantification method is also used as an element item in the scoring item, so that the rating quantification method can be directly used to determine the score of the scoring item in the future. The rating quantification method can be obtained by parsing the rating quantification method file corresponding to the rating method, and the rating quantification method file includes the rating quantification method of each scoring item in the object rating method, wherein the parsing process can be implemented through the above steps S11-S13, which will not be repeated here. In addition, after obtaining the rating quantification method, the rating quantification method can be input into the preset large language model, and the code function and quantitative index corresponding to the rating quantification method can be output through the preset large language model to obtain the quantitative data of the scoring item.

[0078] It should be noted that when determining the several scoring items and scoring data of the scoring items corresponding to the rating method, the scoring item name can be determined first, and then the scoring basis, scoring value and quantitative data can be determined, or the scoring item name can be determined first, and then the quantitative data, scoring basis and scoring value can be determined. In other words, the order of obtaining the scoring basis, scoring value and quantitative data can be adjusted according to user needs, and no specific restrictions are made here.

[0079] The embodiment of the present application forms a rating method set by parsing the existing rating report, and each rating method in the rating method set corresponds to a number of rating items including rating item names, rating values, rating basis and quantitative data. In this way, the rating method set can be used as a priori knowledge set to provide a reference for the large model through the rating values ​​and rating basis of the rating items, so that the large model can more accurately score the rating items of the rated object, thereby improving the rating accuracy. At the same time, for quantifiable rating items, the rating items can be quantified directly based on the quantitative data, which can improve the calculation speed of the rating of the rating items, and thus improve the efficiency of enterprise object rating.

[0080] S20. Determine the target rating method corresponding to the to-be-rated object and the target scoring item corresponding to the target rating method according to the rating method set.

[0081] Specifically, the target rating method is included in the rating method set, that is, when it is necessary to rate the object to be rated, a target rating method can be selected for the object to be rated in the rating method set. Among them, when selecting the target rating method for the object to be rated, the target rating method can be selected for the object to be rated according to the industry in which the object to be rated is located, or a prompt word can be constructed for the object to be rated, and then the target rating method can be selected for the object to be rated based on the prompt word and the object name of the object to be rated through a large model, or the target rating method can be selected by string matching or embedding similarity matching between the corporate object profile of the object to be rated and the instructions for use of the rating method. In addition, after obtaining the target rating method, all the rating items included in the target rating method can be read, and the quantitative category of each rating item can be determined based on the quantitative data of each rating item, so as to determine the scoring acquisition method of the rating item based on the quantitative category of the rating item (including the first quantitative category and the second quantitative category).

[0082] S30: Obtain a first target rating item of a first quantization category in the target rating items, determine first target rating data for the first target rating item, and score the to-be-rated object based on the first target rating data and a preset large language model to determine a first rating.

[0083] Specifically, the first quantitative category indicates that the first score of the first target scoring item cannot be obtained through quantitative calculation, for example, the first target scoring item is the regional layout of branches, etc. For the first target scoring item of the first quantitative category, the first target scoring data corresponding to the first target scoring item included in the rating method set can be used as a reference, and the first score of the first target scoring item can be determined through the large model based on the object data of the object to be rated, wherein the first target scoring data corresponding to the first target scoring item is used to provide prior knowledge for the large model, so that the large model learns the scoring method of the first target scoring item, and then scores the first target scoring item of the object to be rated, which can improve the accuracy of the first score on the one hand, and ensure the consistency and comparability of the scores of the object to be rated and the rated enterprise object on the other hand.

[0084] Further, each rating method includes several rating items, and different rating methods may include the same rating items, so that the first target rating data corresponding to the first target rating item (i.e., the first target rating data corresponding to the first target rating item) can be searched in the several rating items of each rating method (i.e., the search basis is the same or similarity of the rating item name, etc.) to obtain several first target rating data corresponding to the first target rating item, wherein all the first target rating data corresponding to the first target rating item can be included in the existing rating reports of multiple rated enterprise objects. For example, the rating method set includes rating method A and rating method B, rating method A includes rating item A1, rating item A2 and rating item A3, rating method B includes rating item B1, rating item B2 and rating item B3, and rating item A1 and rating item B1 both match the first target rating item, then the first target rating data of the first target rating item is determined based on rating item A1 and rating item B1, for example, the rating basis in rating item A1 and the rating basis in rating item B1 are used as the first target rating data of the first target rating item.

[0085] For example, Figure 3 As shown, the step of scoring the to-be-rated object based on the first target scoring data and the preset large language model to determine the first score specifically includes: S31, obtaining a number of keywords corresponding to the first target rating item; S32, determining summary text about the several keywords based on the object name of the object to be rated and the several keywords; S33, based on the summary text, selecting first target scoring data for the first target scoring item in the rating method set; S34: Based on the summary text and the first target rating data, the object to be rated is rated by using a preset large language model to determine a first rating.

[0086] Specifically, in step S31, several keywords corresponding to the first target scoring item are used as the basis for selecting the summary text of the first target scoring item. That is to say, several keywords are used as search keywords to determine the summary text about several keywords, wherein the summary text includes the operating conditions of the object to be rated, etc.

[0087] Exemplarily, the acquiring of several keywords corresponding to the first target rating item specifically includes: S311, segmenting the scoring basis in the first target scoring data to obtain a plurality of words; S312, respectively calculating the similarity between each word and the scoring item name of the first target scoring item and the word frequency-inverse document frequency of each word; S313, determining the importance of each word based on the similarity of each word and the word frequency-inverse file frequency; S314. Selecting a number of keywords from a number of words based on the importance of each word to obtain a number of keywords corresponding to the first target scoring item.

[0088] Specifically, the similarity is used to reflect the similarity between a word and the first target scoring item, wherein the similarity can be the cosine similarity between the word embedding of the word and the word embedding of the first target scoring item. The term frequency-inverse file frequency is used to reflect the importance of the word to all scoring criteria of the first target scoring item, wherein the term frequency refers to the ratio of the number of occurrences of the word in all scoring criteria of the first target scoring item to the total number of terms in all scoring criteria of the first target scoring item, and the inverse file frequency refers to the logarithm of the ratio of the number of scoring criteria in which the word appears to the total number of all scoring criteria, and the term frequency-inverse file frequency is equal to the product of the term frequency and the inverse file frequency.

[0089] After obtaining the similarity of the words and the word frequency-inverse file frequency, the product of the similarity and the word frequency-inverse file frequency can be used as the importance of the words, and then a number of words are selected in descending order of importance as the keywords of the first target scoring item. Of course, in actual applications, the similarity and the word frequency-inverse file frequency can also be weighted to determine the importance of the words.

[0090] The embodiment of the present application selects several keywords for the first target scoring item based on importance, and then can retrieve a summary text used as a basis for rating the object to be rated based on the several keywords, thereby improving the explanatory power of the summary text for the scoring item, thereby highlighting the scoring basis used for scoring.

[0091] It should be noted that several keywords can also be determined in other ways. For example, they can be set manually, or the importance can be directly determined by taking the similarity between each word and the scoring item name of the first target scoring item as the importance, or the importance can be directly determined by taking the word frequency-inverse file frequency as the importance, etc.

[0092] Further, in step S32, the summary text is a text material used to reflect the status of the enterprise object of the object to be rated, for example, the summary text may include text material reflecting the business status of the object to be rated, etc. The summary material may be obtained by searching with a number of keywords as search keywords, for example, a local database material search may be performed with a number of keywords as search keywords, or a network search may be performed with a number of keywords as search keywords, or a local database material and a network search may be performed simultaneously with a number of keywords as search keywords.

[0093] Exemplarily, the determining of summary texts about the keywords based on the object name of the object to be rated and the keywords specifically includes: S321, searching for text resources of the object to be rated based on the object name and several keywords of the object to be rated; S322, dividing the searched text resources into a number of natural segments, and selecting a preset number of representative natural segments from the natural segments based on the object name and a number of keywords; S323: Based on the preset number of representative natural paragraphs and the keywords, determine summary texts about the keywords through a large language model.

[0094] Specifically, the object name of the object to be rated and several keywords are used as search keywords to search for text resources about the object to be rated that include several keywords, and the text material is used as the basis for scoring the first target scoring item of the object to be rated. The text resource search process can adopt a word matching method or a word embedding similarity method.

[0095] After obtaining the text data, the text data can be divided into several natural segments, and then the natural segments can be screened based on the similarity between the natural segments and several keywords. Specifically, the keywords and the object name of the object to be rated can be concatenated to form a sentence, and then the similarity between the sentence embedding of each natural segment and the sentence embedding of the sentence is calculated (such as cosine similarity, etc.), and then a preset number of representative natural segments are selected in descending order of similarity. Of course, in actual applications, several target natural segments can also be selected based on the number of keywords included in the natural segment.

[0096] After obtaining a preset number of representative natural segments, the prompting project can be used to use the big model to summarize around several keywords based on several target natural segments to obtain a summary text, wherein the prompting words used in the prompting project are used to instruct the big model to summarize around several keywords, for example, the prompting words can be "Please summarize around the following keywords based on the above information." That is to say, when determining the summary text, a summary prompting word can be first constructed to instruct the big model to summarize around several keywords, and then the summary prompting word, several target natural segments and several keywords are spliced, and then input into the big model to output the summary text through the big model.

[0097] The present application screens the textual materials of the object to be rated, and then forms a summary text around several keywords based on the screened target natural paragraphs, so that the summary text can reflect the corporate object status (for example, operating status, etc.) of the object to be rated. In this way, when determining the score of the first target scoring item, the scoring basis and score of the rated corporate object with a similar corporate object status to the object to be rated can be selected based on the summary text as a reference to determine the score of the first target scoring item. This can not only improve the scoring accuracy, but also ensure the consistency and comparability of the scores of the object to be rated and the rated corporate object.

[0098] Of course, in actual applications, the found text material can be directly used as the summary text, or several selected target paragraphs can be used as the summary text, etc.

[0099] Further, in step S33, a plurality of target scoring data are selected from the first target scoring data corresponding to the first target scoring item based on the summary text, and are used as prior knowledge for determining the first score, so that the large model can determine the score corresponding to the summary text based on the prior knowledge. When selecting a plurality of target scoring data from the first target scoring data corresponding to the first target scoring item, the selection may be based on the similarity between the scoring basis in the scoring data and the summary text, or may be randomly selected, etc.

[0100] In an embodiment of the present application, after selecting all first target scoring data corresponding to the first target scoring item, the similarity (such as cosine similarity, etc.) between the text embedding of the summary text and the text embedding of the scoring basis in each first target scoring data is calculated, and then several target scoring data are selected in order of similarity from high to low to obtain several target scoring data that are most similar to the summary text.

[0101] Further, in step S33, after obtaining a plurality of target rating data, the rating basis and rating in each target rating data are extracted, and the first rating of the first target rating item is determined by the large model using the prompt engineering. Exemplarily, the first rating of the object to be rated is rated by the preset large language model based on the summary text and the first target rating data, and the determination of the first rating specifically includes: Constructing a scoring prompt word for the first target scoring item; The scoring prompt word, the first target scoring item, the first target scoring data of the first target scoring item and the summary text are spliced ​​and input into a preset large language model, and the first score of the first target scoring item is output through the preset large language model.

[0102] Specifically, the scoring prompt is used to prompt the large model to output the first score of the first target scoring item. For example, the scoring prompt is "The above are descriptions of some enterprise objects about {fill in the scoring item name} and the corresponding scores. Please use this as a reference to score the enterprise objects described below." After determining the scoring prompt, the scoring basis and score in each target scoring data are spliced ​​as an example, and then the scoring prompt and the summary text are spliced ​​and input into the large model, and the first score of the first target scoring item is output through the large model.

[0103] It should be noted that, in actual applications, the object to be rated, the first target rating item and the first target rating data corresponding to the first target rating item can also be directly used as input items of the big model, and the enterprise object information of the object to be rated can be obtained through the big model, and the score of the first target rating item can be determined based on the obtained enterprise object information, the first target rating item and the first target rating data corresponding to the first target rating item, wherein the score of the first target rating item given for the enterprise object information of the object to be rated, that is, the score of the first target rating item is a score of the object to be rated.

[0104] S40: Acquire a second target scoring item of a second quantitative category in the target scoring items, determine second target scoring data for the second target scoring item, and score the object to be rated based on the second target scoring data to determine a second score.

[0105] Specifically, the second target scoring item of the second quantization category is a scoring item for which the second score can be directly calculated by the rating quantization method. That is to say, for the second target scoring item of the second quantization category, the rating quantization method corresponding to the second target scoring item can be directly determined, and then the second score of the second target scoring item is calculated by the rating quantization method. To this end, for the second target scoring item of the second quantization category, the second target scoring data corresponding to the second target scoring item can be directly selected from the rating method set, and then the second score of the second target scoring item is calculated based on the second target scoring data. In other words, the second target scoring data can be quantitative data. Of course, the second target scoring data can also include the scoring item name, the rating value, the scoring basis and the quantitative data corresponding to the scoring item, and then the quantitative data is read in the second target scoring data. In addition, when there are multiple second target scoring data corresponding to the second target scoring item, one second target scoring data can be randomly selected from the multiple second target scoring data, and then the quantitative data in the second target scoring data is used as the quantitative data for determining the second score of the second target scoring item.

[0106] For example, Figure 4 As shown, the step of scoring the object to be rated based on the second target rating data and determining the second rating specifically includes: S41, obtaining quantitative data in the second target scoring data, wherein the quantitative data includes a code function corresponding to a rating quantitative method and a quantitative index required by the rating quantitative method; S42: Use the quantitative indicator data as input of the code function, and execute the code function to obtain a second score for the second target scoring item.

[0107] Specifically, the quantitative indicator may be a financial indicator, an enterprise object scale indicator, etc., used to calculate the second score. After the quantitative indicator is read, the corresponding quantitative indicator data of the quantitative indicator will be queried from the enterprise object data of the object to be rated, and then the quantitative indicator data will be used as the input of the code function, and the second score of the second target scoring item can be obtained by executing the code function, wherein the code function can be a Python function, etc.

[0108] The present application constructs a code function in advance, and then after obtaining the quantitative indicator data, the quantitative indicator data is input into the code function to quickly calculate the second score of the second target scoring item, so that the scoring process is simple and accurate results can be obtained.

[0109] S50: Determine a rating level of the object to be rated according to the first rating and the second rating.

[0110] Specifically, after obtaining all the first scores and all the second scores, the first scores and the second scores can be directly weighted and summed to obtain the final score of the enterprise to be rated, and finally the rating level of the object to be rated can be determined according to the corresponding relationship between the final score and the enterprise object rating. Of course, the average score of all the first scores and all the second scores can also be used as the final score of the enterprise to be rated.

[0111] Exemplarily, determining the rating level of the to-be-rated object according to the first rating and the second rating specifically includes: Weighting the first score and the second score to obtain a final score; According to the preset correspondence between the rating level and the rating, the rating level corresponding to the final rating is determined to obtain the rating level of the object to be rated.

[0112] Specifically, a weight coefficient can be configured for each scoring item in advance, and then the score (first score or second score) corresponding to each scoring item can be weighted based on the weight coefficient of each scoring item. This can adjust the weight coefficient of each scoring item according to actual needs, so that the final score is more accurate. The correspondence between the scoring level and the enterprise object score is pre-set. For example, several scoring levels are pre-set, and then the scoring interval corresponding to each scoring level is set to form a correspondence between the scoring level and the enterprise object score. Of course, the correspondence between the scoring level and the enterprise object score can also be adjusted according to actual conditions.

[0113] In summary, this embodiment provides an object rating method, the method comprising obtaining a preset rating method set; determining a target rating method corresponding to the object to be rated and a target scoring item corresponding to the target rating method according to the rating method set; obtaining a first target scoring item of a first quantitative category in the target scoring item, determining a first target scoring data of the first target scoring item, and scoring the object to be rated based on the first target scoring data and a preset large language model to determine a first score; for a quantifiable second target scoring item in the object rating method, determining a second score of the second target scoring item based on the quantitative data in the second target scoring data corresponding to the second target scoring item; and determining the rating level of the object to be rated according to the first score and the second score. This application uses a rating method set formed based on an existing rating report of an already rated enterprise object as prior knowledge, determines the score of the non-quantifiable scoring item corresponding to the object to be rated through a large model, and directly calculates the score of the quantifiable scoring item through the quantitative data in the rating method set, which can not only reduce the labor cost and time cost required for manual rating, but also save the time cost spent in the enterprise object information screening process of the object to be rated. At the same time, by using the rating method set as prior knowledge, the accuracy of enterprise object ratings and the consistency of ratings with rated enterprise objects can also be improved.

[0114] Based on the above object rating method, this embodiment provides an object rating device, such as Figure 5 As shown, the object rating device specifically includes: A construction module 100 is used to construct a rating method set based on an existing rating report, wherein the rating method set includes a plurality of rating methods and a plurality of rating items corresponding to each rating method, wherein the rating items include a plurality of rating data target rating methods; A rating method determination module 200 is used to determine a target rating method corresponding to the to-be-rated object and a target scoring item corresponding to the target rating method according to the rating method set; The first processing module 300 is used to obtain a first target rating item of a first quantization category in the target rating item, determine first target rating data for the first target rating item, and rate the to-be-rated object based on the first target rating data and a preset large language model to determine a first rating; The second processing module 400 is used to obtain a second target scoring item of a second quantitative category in the target scoring item, determine second target scoring data for the second target scoring item, and score the to-be-rated object based on the second target scoring data to determine a second score; The grade determination module 500 is used to determine the grade of the object to be rated according to the first score and the second score.

[0115] Based on the above object rating method, this embodiment provides a computer-readable storage medium, which stores one or more programs. The one or more programs can be executed by one or more processors to implement the steps in the object rating method as described in the above embodiment.

[0116] Based on the above object rating method, the present application also provides a terminal device, such as Figure 6 As shown, it includes at least one processor 20; a display screen 21; and a memory 22, and may also include a communications interface 23 and a bus 24. The processor 20, the display screen 21, the memory 22, and the communications interface 23 may communicate with each other through the bus 24. The display screen 21 is configured to display a preset user guide interface in the initial setting mode. The communications interface 23 may transmit information. The processor 20 may call the logic instructions in the memory 22 to execute the method in the above embodiment.

[0117] In addition, the logic instructions in the memory 22 can be implemented in the form of software functional units and can be stored in a computer-readable storage medium when sold or used as an independent product.

[0118] The memory 22 is a computer-readable storage medium that can be configured to store software programs, computer executable programs, such as program instructions or modules corresponding to the methods in the embodiments of the present disclosure. The processor 20 executes functional applications and data processing by running the software programs, instructions or modules stored in the memory 22, that is, implementing the methods in the above embodiments.

[0119] The memory 22 may include a program storage area and a data storage area, wherein the program storage area may store an operating system and at least one application required for a function; the data storage area may store data created according to the use of the terminal device, etc. In addition, the memory 22 may include a high-speed random access memory and may also include a non-volatile memory. For example, a variety of media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a disk or an optical disk, may also be a transient storage medium.

[0120] In addition, the specific process of loading and executing the multiple instruction processors in the above-mentioned storage medium and the terminal device has been described in detail in the above-mentioned method, and will not be described one by one here.

[0121] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit it. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. An object rating method, characterized in that: The object rating method specifically includes: Building a rating method set based on an existing rating report, the rating method set including a number of rating methods and a number of scoring items corresponding to each rating method, the scoring items including a plurality of scoring data; Determine the target rating method corresponding to the to-be-rated object and the target scoring item corresponding to the target rating method according to the rating method set; Acquire a first target scoring item of a first quantization category in the target scoring items, determine first target scoring data for the first target scoring item, and score the to-be-rated object based on the first target scoring data and a preset large language model to determine a first score; Acquire a second target scoring item of a second quantitative category in the target scoring item, determine second target scoring data for the second target scoring item, and score the to-be-rated object based on the second target scoring data to determine a second score; The rating level of the object to be rated is determined according to the first rating and the second rating.

2. The object rating method according to claim 1, characterized in that: The rating method set constructed based on the existing rating report specifically includes: Obtain several existing rating reports; Respectively obtain the detection target of each existing rating report and the detection data corresponding to the detection target, wherein the detection target includes a title, a paragraph and a table, and the detection data includes category information, location information and text content; Determine, based on the detection data of all detection targets of each existing rating report, a rating method adopted by each existing rating report, a number of rating items corresponding to the rating method, and a plurality of rating data of the rating items; A rating method set is constructed according to the determined rating methods, a number of rating items corresponding to each rating method, and a plurality of rating data of the rating items.

3. The object rating method according to claim 2, characterized in that: The obtaining of the detection target of each existing rating report and the detection data corresponding to the detection target specifically includes: Input each of the existing rating reports into a trained text layout detection model, and output the detection targets in each of the existing rating reports and the category information and location information of the detection targets through the text layout detection model; Extracting text content of each of the detection targets from the corresponding existing rating reports based on the location information of each of the detection targets; The category information, the location information and the text content are used as detection data of the detection target, so as to obtain the detection target of each existing rating report and the detection data corresponding to the detection target.

4. The object rating method according to claim 3, characterized in that: The trained text layout detection model specifically includes: Acquire a text layout dataset, wherein the text layout dataset includes a plurality of training texts and annotation target information corresponding to the training texts; The preset target detection model is trained using the training text in the text layout data to obtain a trained text layout detection model.

5. The object rating method according to claim 2, characterized in that: The step of determining the rating method adopted by each existing rating report, the multiple scoring items corresponding to the rating method, and the multiple scoring data of the scoring items based on the detection data of all detection targets of each existing rating report specifically includes: Determining the rating method adopted by the existing rating report based on the detection data of the natural segment; Based on the detection data of the detection target, several scoring items and multiple scoring data of the scoring items corresponding to the rating method are determined.

6. The object rating method according to claim 5, characterized in that: The scoring data includes the scoring item name, scoring value, scoring basis and quantitative data. The detection data based on the detection target determines the several scoring items and scoring data of the scoring items corresponding to the rating method, specifically including: Determine a number of scoring items corresponding to the rating method and the scoring item name and scoring value of each scoring item based on the detection data in the table; Determine the scoring basis of each scoring item based on the detection data of the title and the natural paragraph; The rating quantification method corresponding to each rating item is obtained, and the quantitative data of each rating item is determined through a preset large language model based on the rating quantification method.

7. The object rating method according to claim 6, characterized in that: The scoring basis for determining each scoring item based on the detection data of the title and the natural paragraph specifically includes: Determine the target title corresponding to each rating item, and obtain the target paragraph corresponding to the target title; Based on the text content in the detection data of the target title and the text content in the detection data of the target paragraph, a scoring basis for the scoring item name is determined.

8. The object rating method according to claim 7, characterized in that: The step of determining the target title corresponding to each rating item and obtaining the target segment corresponding to the target title specifically includes: Obtaining a structured directory of the existing rating report, wherein the structured directory consists of multiple levels of titles and contents contained in each level of titles; Determine the target title corresponding to each scoring item according to the scoring item name of each scoring item and the structured directory; The target paragraph corresponding to the target title is determined according to the report content contained in the target title.

9. The object rating method according to claim 8, characterized in that: The process of constructing the structured directory specifically includes: Determine the title level of each title in the existing rating report, and determine the subordinate relationship between the titles and the content contained in each title based on the title level of each title and the position information of each title; Based on the subordinate relationships between the titles and the content contained in each title, a structured directory of the existing rating reports is constructed.

10. The object rating method according to claim 9, characterized in that: Determining the title level of each title in the existing rating report specifically includes: Get the title font size and title sequence number of each title; Based on the obtained font size and title sequence number of each title, the title level of each title is determined.

11. The object rating method according to claim 9, characterized in that: The process of determining the content contained in each title specifically includes: Based on the position information of each detection target, the detection targets are sorted according to the order in the existing rating report to obtain a detection target sequence; The detection target and detection data between two adjacent titles of the same level in the detection target sequence are taken as the contents contained in the title sorted first.

12. The object rating method according to claim 1, characterized in that: Scoring the to-be-rated object based on the first target scoring data and the preset large language model, and determining the first score specifically includes: Obtaining several keywords corresponding to the first target rating item; Determining summary texts about the keywords based on the object name of the object to be rated and the keywords; Based on the summary text, selecting first target scoring data for the first target scoring item in the rating method set; Based on the summary text and the first target rating data, the to-be-rated object is rated by using a preset large language model to determine a first rating.

13. The object rating method according to claim 12, characterized in that: The step of scoring the object to be rated by using a preset large language model based on the summary text and the first target rating data to determine the first rating specifically includes: Constructing a scoring prompt word for the first target scoring item; The scoring prompt word, the first target scoring item, the first target scoring data of the first target scoring item and the summary text are spliced ​​and input into a preset large language model, and the first score of the first target scoring item is output through the preset large language model.

14. The object rating method according to claim 12, characterized in that: The step of obtaining a number of keywords corresponding to the first target rating item specifically includes: Segmenting the scoring basis in the first target scoring data to obtain a plurality of words; Calculate the similarity between each word and the scoring item name of the first target scoring item and the word frequency-inverse document frequency of each word respectively; Determining the importance of each word based on the similarity of each word and the word frequency-inverse document frequency; A number of keywords are selected from the number of words based on the importance of each word to obtain a number of keywords corresponding to the first target scoring item.

15. The object rating method according to claim 12, characterized in that: The step of determining summary texts about the keywords based on the object name of the object to be rated and the keywords specifically includes: Searching for text resources of the object to be rated based on the object name and a number of keywords of the object to be rated; Dividing the searched text resources into a number of natural segments, and selecting a preset number of representative natural segments from the natural segments based on the object name and a number of keywords; Based on the preset number of representative natural paragraphs and the several keywords, summary texts about the several keywords are determined by using a preset large language model.

16. The object rating method according to claim 1, characterized in that: Scoring the to-be-rated object based on the second target scoring data and determining the second score specifically includes: Acquire quantitative data in the second target scoring data, wherein the quantitative data includes a code function corresponding to a rating quantification method and a quantitative index required by the rating quantification method; Obtaining quantitative indicator data corresponding to the quantitative indicator; The quantitative indicator data is used as an input of the code function to obtain a second score for the second target scoring item.

17. The object rating method according to claim 1, characterized in that: Determining the rating level of the to-be-rated object according to the first rating and the second rating specifically includes: Weighting the first score and the second score to obtain a final score; According to the preset correspondence between the rating level and the rating, the rating level corresponding to the final rating is determined to obtain the rating level of the object to be rated.

18. An object rating device, characterized in that: The object rating device specifically comprises: A construction module, used to construct a rating method set based on an existing rating report, wherein the rating method set includes a plurality of rating methods and a plurality of scoring items corresponding to each rating method, wherein the scoring items include a plurality of rating data target rating methods; A rating method determination module, used to determine a target rating method corresponding to the to-be-rated object and a target scoring item corresponding to the target rating method according to the rating method set; A first processing module is used to obtain a first target rating item of a first quantization category in the target rating item, determine first target rating data for the first target rating item, and score the to-be-rated object based on the first target rating data and a preset large language model to determine a first rating; A second processing module is used to obtain a second target scoring item of a second quantitative category in the target scoring item, determine second target scoring data for the second target scoring item, and score the to-be-rated object based on the second target scoring data to determine a second score; A grade determination module is used to determine the grade of the object to be rated according to the first score and the second score.

19. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores one or more programs, and the one or more programs can be executed by one or more processors to implement the steps in the object rating method according to any one of claims 1-17.

20. A terminal device, characterized in that: include: Processor and memory; The memory stores a computer-readable program executable by the processor; When the processor executes the computer-readable program, the steps of the object rating method according to any one of claims 1 to 17 are implemented.

Images