Rating feature classification method for rating model of science and technology enterprise

Abstract

The invention relates to the field of feature analysis, in particular to a rating feature classification method for a rating model of a science and technology enterprise, which comprises the following steps: setting a retrieval target, acquiring a plurality of retrieval interfaces associated with the retrieval target, performing interface visual analysis on the retrieval interfaces, and determining a visual clustering region based on the relative position of target features. Calculating relevance parameters of target features and retrieval targets, setting classification labels for the visual clustering areas, and performing adaptive selection analysis on the target features in the visual clustering areas according to the labels, including selecting and removing the target features in the fuzzy areas of the retrieval targets, and selecting and removing the target features in the fuzzy areas of the retrieval targets; according to the method, features highly related to the retrieval target are reserved or all target features in the visual clustering region are removed, and the reserved target features are subsequently transmitted to a preset language rating model, so that irrelevant features in the webpage can be effectively reduced, the efficiency and accuracy of obtaining the features from the webpage are improved, and the user experience is improved. And the feedback accuracy of the language rating model is ensured.

Description

Technical Field

[0001] The present invention relates to the field of feature analysis, and particularly to a rating feature classification method for a rating model of technology enterprises. Background Art

[0002] With the development of computer science, large natural language models have emerged. By inputting content into a large natural language model, it can analyze the content and give corresponding feedback. Currently, some natural language models already have the ability to extract information from web pages and then give comprehensive feedback based on the extracted information. Based on this, large natural language models are widely used in various fields, especially in the enterprise consulting field. Relevant consultations about enterprises can be put forward to the large natural language model, and then comprehensive feedback can be given based on the large natural language model to provide support for the consultants.

[0003] For example, Chinese Patent Publication No.: CN118861222A discloses a consultation processing method, device, storage medium, and computer device based on a large language model. The method includes: when it is outside the working hours of the artificial customer service, receiving the target consultation text input by the user based on the customer service function; constructing a target intention analysis prompt word according to the target consultation text and a preset intention analysis prompt word template, and calling the large language model to generate the consultation intention corresponding to the target consultation text according to the target intention analysis prompt word; determining the target processing method for the target consultation text according to the consultation intention, and performing consultation processing according to the target processing method.

[0004] However, there are still the following problems in the prior art.

[0005] In the prior art, when the user side puts forward consultation information and the large model gives feedback based on the information in the web page, due to the large number of web page interfaces and the mixed content, the relevant features in the web page are extracted slowly, and irrelevant information may be introduced to interfere with the feedback of the large model, resulting in low feedback efficiency and poor feedback results. Summary of the Invention

[0006] Therefore, the present invention provides a rating feature method for a rating model of technology enterprises to overcome the problems in the prior art that when the user side puts forward consultation information and the large model gives feedback based on the information in the web page, due to the large number of web page interfaces and the mixed content, the relevant features in the web page are extracted slowly, and irrelevant information may be introduced to interfere with the feedback of the large model, resulting in low feedback efficiency and poor feedback results.

[0007] To achieve the above object, the present invention provides a rating feature classification method for a rating model of technology enterprises, which includes:

[0008] Setting a retrieval target and obtaining a number of retrieval interfaces associated with the retrieval target;

[0009] Perform interface visual analysis on the retrieval interface, including identifying several target features in the page, clustering the target features based on the relative position relationship between the target features, and determining several visual clustering regions;

[0010] Extract some target features in the visual clustering region, determine the relevance parameter between the target features and the retrieval target, and set classification labels for each of the visual clustering regions;

[0011] Analyze and extract the target features in each visual clustering region according to the classification label, including,

[0012] Identify the calibration keywords strongly associated with the retrieval target in the visual clustering region, determine the position information of the calibration keywords in the visual clustering region, screen out the calibration keyword groups used to retrieve the fuzzy region position of the target, determine the retrieval target fuzzy region in the visual clustering region, remove the target features in the retrieval target fuzzy region, and retain the remaining target features;

[0013] Or, remove all target features in the visual clustering region;

[0014] Transmit the retained target features to a preset language rating model;

[0015] Among them, the retrieval target is in text form and is set by the user side.

[0016] Further, the process of clustering the target features based on the relative position relationship between the target features and determining several visual clustering regions includes,

[0017] Determine the coordinates of several target features in the retrieval interface;

[0018] Determine the relative distance between the target features based on each of the coordinates;

[0019] Cluster each target feature based on the relative distance to obtain several clustering clusters;

[0020] Determine each edge target feature in the clustering cluster, and connect each of the edge target features in sequence to obtain a connection region;

[0021] Enlarge the connection region by a predetermined multiple to obtain the visual clustering region;

[0022] Among them, the relative distance between the target features in each clustering cluster and the nearest target feature is less than a predetermined distance threshold.

[0023] Further, the process of extracting some target features in the visual clustering region includes,

[0024] Divide the text composed of the target features in the visual clustering region into several text segments;

[0025] Extract a number of text segments at a predetermined extraction interval.

[0026] Furthermore, the process of determining the relevance parameter between the target feature and the retrieval target includes,

[0027] Determine the text segments corresponding to the extracted partial target features in the visual clustering region;

[0028] Compare the text segments with the retrieval target respectively to determine the semantic relevance degree corresponding to each text segment, and determine the average value of the semantic relevance degrees as the relevance parameter.

[0029] Furthermore, setting classification labels for each visual clustering region includes,

[0030] If the relevance parameter is greater than or equal to a predetermined relevance degree standard threshold, set an association label for the visual clustering region;

[0031] If the relevance parameter is less than the predetermined relevance degree standard threshold, set a fuzzy label for the visual clustering region.

[0032] Furthermore, analyzing and extracting the target features in each visual clustering region according to the classification labels includes,

[0033] If the visual clustering region is an association label, identify the calibration keywords strongly associated with the retrieval target in the visual clustering region, determine the position information of the calibration keywords in the visual clustering region, so as to screen the calibration keyword group for retrieving the fuzzy region position of the target, determine the fuzzy region of the retrieval target in the visual clustering region, remove the target features in the fuzzy region of the retrieval target, and retain the remaining target features;

[0034] If the visual clustering region is a fuzzy label, remove all the target features in the visual clustering region.

[0035] Furthermore, the process of screening the calibration keyword group for referring to the fuzzy region spacing includes,

[0036] Based on the position information of the calibration keywords, determine the distance between each calibration keyword and the nearest adjacent calibration keyword in the reference direction;

[0037] If there is a distance between a calibration keyword and the nearest adjacent calibration keyword greater than a predetermined fuzzy region spacing threshold;

[0038] Then determine the calibration keyword and the nearest adjacent calibration keyword as the calibration keyword group;

[0039] Wherein, the reference direction is the direction perpendicular to the bottom edge of the retrieval interface.

[0040] Further, the process of determining the retrieval target fuzzy area in the visual clustering area includes

[0041] Construct a reference line parallel to the reference direction at the center of the visual clustering area;

[0042] Construct a calibration line perpendicular to the reference line on the reference line through the positions of the keywords in the calibrated keyword group.

[0043] Determine the area enclosed by each calibration line and the edge of the visual clustering area as the retrieval target fuzzy area.

[0044] Further, obtain a number of retrieval interfaces associated with the retrieval target, where

[0045] If the semantic association degree between the title corresponding to the retrieval interface and the retrieval target is greater than a predetermined association degree standard threshold, it is determined that the retrieval interface is associated with the retrieval target.

[0046] Further, the target features include text features and symbol features.

[0047] Compared with the prior art, the beneficial effects of the present invention are as follows. By setting a retrieval target, obtaining a number of retrieval interfaces associated with the retrieval target, performing interface visual analysis on the retrieval interfaces, determining a visual clustering area based on the relative positions of the target features, calculating the relevance parameters between the target features and the retrieval target, setting classification labels for each visual clustering area, and performing adaptive selection analysis on the target features in the visual clustering area according to the labels, including selecting and removing the target features in the retrieval target fuzzy area, retaining the features with a relatively high relevance to the retrieval target or removing all the target features in the visual clustering area, and then transmitting the retained target features to a preset language rating model. The present invention can effectively reduce the irrelevant features in the web page, improve the efficiency and accuracy of obtaining features from the web page, and ensure the accuracy of the feedback of the language rating model.

[0048] In particular, the present invention determines the visual clustering region based on the relative position of the target features by performing interface visual analysis on the retrieval interface. In actual situations, the distributions of different retrieval interfaces vary. Only based on the title, it is impossible to determine the relevance between the content of the entire retrieval interface and the detection target. Different retrieval interfaces may have various information regions and other layout situations, and similar content usually clusters relatively. Therefore, performing language analysis on the text in the entire page will waste a large amount of computing power when facing a large number of interfaces. Therefore, it is considered to observe the clustering of target features in the page from the visual dimension to determine the visual clustering region, without directly adopting the semantic association analysis method. Clustering analysis from the image dimension is relatively simple, which improves the analysis efficiency. For a single visual clustering region, only part of the target features are extracted to determine the relevance parameter with the retrieval target, and labels are set for the visual clustering region, reducing the amount of data analysis. Therefore, the present invention determines the visual clustering region and sets classification labels, which can dynamically adapt to different retrieval interface layouts, can effectively identify and divide the visual clustering region. Furthermore, when facing a large number of retrieval interfaces, it can effectively reduce the irrelevant features in the web page, improve the efficiency and accuracy of obtaining features from the web page, and ensure the accuracy of the feedback of the language rating model.

[0049] In particular, the present invention adaptively adopts different analysis and extraction methods for the visual clustering regions with different classification labels in the retrieval interface. For the associated labels, considering the effectiveness of the text in the entire visual clustering region, the position of the calibration keywords is determined, and then the calibration keyword group is determined to further determine the fuzzy region of the retrieval target. In actual situations, the content corresponding to the text in the visual clustering region is not all relevant to the retrieval target. Usually, if the text is the main text describing the retrieval target, calibration keywords strongly associated with the retrieval target will be relatively evenly distributed in the text. If there is content irrelevant to the retrieval target in the visual clustering region, there may be no calibration keywords strongly associated with the retrieval target in this part of the content. Therefore, for the special case where the calibration keywords are widely spaced, the fuzzy region of the retrieval target is determined, and the target features therein are removed. Furthermore, when facing a large number of retrieval interfaces, it can effectively reduce the irrelevant features in the web page, improve the efficiency and accuracy of obtaining features from the web page, and ensure the accuracy of the feedback of the language rating model.

[0050] In particular, for the visual clustering region with fuzzy labels, all the target features therein are directly removed. Finally, the remaining target features are transmitted to the preset language rating model, which can effectively filter out irrelevant information and improve the efficiency and accuracy of information acquisition. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] Figure 1 It is a schematic diagram of the steps of the rating feature classification method for the rating model of technology enterprises in the embodiment of the invention;

[0052] Figure 2 It is a logical decision diagram for identifying the current model area label category in the invention embodiment;

[0053] Figure 3 It is a logical block diagram for analyzing the visual clustering area according to the area label in the invention embodiment;

[0054] Figure 4 It is a logical decision diagram for screening calibration keyword groups for reference to the fuzzy area spacing in the invention embodiment;

[0055] Figure 5 It is a logical block diagram for determining the association between the retrieval interface and the retrieval target in the invention embodiment. Detailed implementation manners

[0056] In order to make the purpose and advantages of the present invention clearer, the present invention will be further described below in conjunction with embodiments; it should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0057] The preferred embodiments of the present invention will be described below with reference to the accompanying drawings. Those skilled in the art should understand that these embodiments are only used to explain the technical principle of the present invention and do not limit the protection scope of the present invention.

[0058] Please refer to Figure 1 As shown, it is a step schematic diagram of a rating feature classification method for a rating model of technology enterprises in an embodiment of the present invention. The rating feature classification method for a rating model of technology enterprises in an embodiment of the present invention includes:

[0059] Step S1, set a retrieval target and obtain a number of retrieval interfaces associated with the retrieval target;

[0060] Step S2, perform interface visual analysis on the retrieval interface, including identifying a number of target features in the page, clustering the target features based on the relative position relationship between the target features, and determining a number of visual clustering areas;

[0061] Step S3, extract some target features in the visual clustering area, determine the relevance parameter between the target feature and the retrieval target, and set classification labels for each of the visual clustering areas;

[0062] Step S4, analyze and extract the target features in each visual clustering area according to the classification label, including

[0063] Identify the calibration keywords strongly associated with the retrieval target in the visual clustering region, determine the position information of the calibration keywords in the visual clustering region, so as to screen the calibration keyword group for retrieving the position of the fuzzy region of the target, determine the fuzzy region of the retrieval target in the visual clustering region, remove the target features in the fuzzy region of the retrieval target, and retain the remaining target features;

[0064] Or, remove all target features in the visual clustering region;

[0065] Step S5, transmit the retained target features to a preset language rating model;

[0066] Wherein, the retrieval target is in text form and is set by the user terminal.

[0067] Specifically, the user terminal can preset the required retrieval target. For example, if it is necessary to obtain the evaluation information of a certain enterprise, the retrieval target can be set as "how about a certain enterprise in certain aspects". Those skilled in the art can set the retrieval target according to the needs, which will not be elaborated here.

[0068] Specifically, the specific form of the language rating model is not limited. Those skilled in the art can adopt the existing open-source natural language large model under the premise of authorization, as long as it can make a feedback based on the consultation content input by the user terminal and the retrieved interface information extracted, which will not be elaborated here.

[0069] Specifically, the method for identifying target features is not limited. In practice, the target features are text features and symbol features. For example, an OCR model can be used to identify text features, and the coordinates of the text features can be located after calibrating the text features. Of course, other methods can also be used, which will not be elaborated here.

[0070] Specifically, the process of clustering the target features based on the relative position relationship between the target features to determine several visual clustering regions includes,

[0071] Determine the coordinates of several target features on the retrieval interface;

[0072] Determine the relative distance between the target features based on each coordinate;

[0073] Cluster each target feature based on the relative distance to obtain several clustering clusters;

[0074] Determine each edge target feature in the clustering cluster, and connect each edge target feature in sequence to obtain a connection region;

[0075] It is understandable that the edge target features in the retrieval interface are target features located at the edge of the area where the clustering clusters are located, and at least one edge condition needs to be satisfied. The edge conditions include the minimum vertical coordinate, the maximum vertical coordinate, the maximum horizontal coordinate, and the minimum horizontal coordinate;

[0076] Enlarge the connection area by a predetermined multiple to obtain the visual clustering area;

[0077] Among them, the relative distance between the target features in each clustering cluster and the nearest target feature is less than a predetermined distance threshold.

[0078] Specifically, the purpose of enlarging by a predetermined multiple is to retain the target features as much as possible without interfering with other visual clustering areas. Preferably, the predetermined multiple is set to 1.1 times.

[0079] Specifically, the predetermined distance threshold is obtained by pre-setting. Among them,

[0080] Pre-obtain a number of retrieval interfaces as samples, determine the average distance between each target feature in the retrieval interface and the nearest target feature, and set 1.15 times the average distance as the predetermined distance threshold.

[0081] Specifically, the process of extracting some target features from the visual clustering area includes,

[0082] Divide the text composed of the target features in the visual clustering area into several text segments;

[0083] Extract several text segments at a predetermined extraction interval.

[0084] In implementation, when dividing the text composed of the target features in the visual clustering area into several text segments, the length of the text segments is not limited. Each text segment can be a sentence, a paragraph, or a text unit divided according to specific rules.

[0085] In implementation, the predetermined extraction interval is selected within the range [2, 5], that is, extract one text segment every text segment at the predetermined extraction interval.

[0086] The present invention sets a retrieval target, obtains a number of retrieval interfaces associated with the retrieval target, performs interface visual analysis on the retrieval interfaces, and determines visual clustering regions based on the relative positions of target features. Calculate the relevance parameter between the target feature and the retrieval target, set classification labels for each of the visual clustering regions, and perform adaptive selection analysis on the target features in the visual clustering regions according to the labels, including selecting and removing the target features in the fuzzy region of the retrieval target, retaining the features with a relatively high relevance to the retrieval target or removing all the target features in the visual clustering region. Subsequently, the retained target features are transmitted to a preset language rating model. The present invention can effectively reduce irrelevant features in the web page, improve the efficiency and accuracy of obtaining features from the web page, and ensure the accuracy of the feedback of the language rating model.

[0087] Specifically, the process of determining the relevance parameter between the target feature and the retrieval target includes,

[0088] Determine the text segments corresponding to some of the target features extracted in the visual clustering region;

[0089] Compare the text segments with the retrieval target respectively to determine the semantic relevance degree corresponding to each text segment, and determine the average value of the semantic relevance degree as the relevance parameter.

[0090] In implementation, the present invention does not limit the method for calculating the semantic relevance degree. The semantic relevance degree between them can be measured by calculating the average value of the cosine similarity between the text segments and the word vectors of the retrieval target. Of course, other methods can also be used, as long as the semantic relevance degrees corresponding to the text segments determined by comparing the text segments corresponding to some of the target features with the retrieval target can be calculated. This will not be elaborated here.

[0091] Please refer to Figure 2 As shown, it is the logical decision diagram for identifying the label categories of the current model region in the embodiment of the invention. Specifically, setting classification labels for each of the visual clustering regions includes,

[0092] If the relevance parameter is greater than or equal to a predetermined relevance standard threshold, set an association label for the visual clustering region;

[0093] If the relevance parameter is less than the predetermined relevance standard threshold, set a fuzzy label for the visual clustering region.

[0094] In implementation, the predetermined relevance standard threshold of the present invention is determined in advance, where,

[0095] Those skilled in the art can collect a large number of corpus samples related to the retrieval target, split the corpus samples into text segments, calculate the average semantic correlation degree between each text segment in the corpus samples and the retrieval target, and set the product of the average semantic correlation degree and the error coefficient as a predetermined correlation degree standard threshold, where the error coefficient is in the interval [0.75, 0.85].

[0096] Please refer to Figure 3 as shown, which is a logic block diagram for analyzing the visual clustering region according to the region label in the invention embodiment. Specifically, analyzing and extracting the target features in each visual clustering region according to the classification label includes,

[0097] If the visual clustering region is an associated label, identify the calibration keywords strongly associated with the retrieval target in the visual clustering region, determine the position information of the calibration keywords in the visual clustering region, so as to screen out the calibration keyword group for the fuzzy region position of the retrieval target, determine the fuzzy region of the retrieval target in the visual clustering region, remove the target features in the fuzzy region of the retrieval target, and retain the remaining target features;

[0098] If the visual clustering region is a fuzzy label, remove all the target features in the visual clustering region.

[0099] The present invention analyzes the interface vision of the retrieval interface, determines the visual clustering region based on the relative position of the target features. In actual situations, the distributions of different retrieval interfaces are different. Only based on the title, it is impossible to determine the relevance between the content of the entire retrieval interface and the detection target. Different retrieval interfaces may have various layout situations such as information regions. Similar contents are usually relatively concentrated. Therefore, for the text in the entire page, performing language analysis will waste a large amount of computing power when facing a large number of interfaces. Therefore, considering observing the aggregation situation of the target features in the page from the visual dimension to determine the visual clustering region, without directly using the semantic correlation analysis method, clustering analysis from the image dimension is relatively simple, which improves the analysis efficiency. For a single visual clustering region, only extract some target features to determine the relevance parameter with the retrieval target, set a label for the visual clustering region, and reduce the amount of data analysis. Therefore, the present invention determines the visual clustering region and sets classification labels, which can dynamically adapt to different retrieval interface layouts, can effectively identify and divide the visual clustering region. Furthermore, when facing a large number of retrieval interfaces, it can effectively reduce the irrelevant features in the web page, improve the efficiency and accuracy of obtaining features from the web page, and ensure the accuracy of the feedback of the language rating model.

[0100] Please refer to Figure 4 as shown, which is a logic decision diagram for screening the calibration keyword group for referring to the fuzzy region spacing in the invention embodiment. Specifically, the process of screening the calibration keyword group for referring to the fuzzy region spacing includes,

[0101] Determine the distance between each calibrated keyword and the nearest calibrated keyword in the reference direction based on the position information of the calibrated keywords;

[0102] If there is a distance between a calibrated keyword and the nearest calibrated keyword that is greater than a predetermined fuzzy region spacing threshold;

[0103] Then determine the calibrated keyword and the nearest calibrated keyword as a calibrated keyword group;

[0104] Wherein, the reference direction is perpendicular to the bottom edge direction of the retrieval interface.

[0105] In implementation, the predetermined fuzzy region spacing threshold of the present invention is determined in advance. Specifically, a number of long texts associated with the retrieval target are obtained in advance, the maximum spacing between the calibrated keywords and the nearest calibrated keyword in each long text is calculated, the average value of the maximum spacings is solved, and the product of the average value of the maximum spacings and the amplification factor is used as the fuzzy region spacing threshold, and the amplification factor is selected within the interval [1.2, 1.5].

[0106] When determining the calibrated keywords strongly associated with the retrieval target, calculate the semantic association degree between each keyword and the retrieval target. If the semantic association degree is greater than the predetermined strong semantic association degree threshold, then determine the keyword as a calibrated keyword.

[0107] In implementation, the strong semantic association degree threshold is determined based on the standard association degree threshold. In implementation, it is set to 1.5 times the standard association degree threshold.

[0108] In implementation, for the association degree between a keyword and the retrieval target, the cosine similarity method is also used. The keyword and several keywords in the retrieval target are both vectorized, the average value of the cosine similarities between the keyword and the keywords in the retrieval target is solved, and the average value of the cosine similarities is determined as the association degree between the keyword and the retrieval target.

[0109] Specifically, the process of determining the retrieval target fuzzy region in the visual clustering region includes,

[0110] Construct a reference line parallel to the reference direction at the center of the visual clustering region;

[0111] Construct a number of calibration lines perpendicular to the reference line through the positions of the keywords in the calibrated keyword group on the reference line.

[0112] Determine the region enclosed by each calibration line and the edge of the visual clustering region as the retrieval target fuzzy region.

[0113] It can be understood that the constructed retrieval target fuzzy region is a rectangular region.

[0114] Specifically, please refer toFigure 5 as shown Figure 5 is a logic block diagram showing the association between the determination search interface and the search target of the invention embodiment. A number of search interfaces associated with the search target are obtained. Among them,

[0115] if the semantic association degree between the title corresponding to the search interface and the search target is greater than a predetermined association degree standard threshold, it is determined that the search interface is associated with the search target.

[0116] Specifically, the target features include text features and symbol features.

[0117] The present invention adaptively adopts different analysis and extraction methods for the visual clustering regions of different classification labels in the search interface. For associated labels, considering the effectiveness of the text in the entire visual clustering region, the position of the calibration keyword is determined, and then the calibration keyword group is determined to further determine the fuzzy region of the search target. In actual situations, the content corresponding to the text in the visual clustering region is not necessarily associated with the search target. Usually, if the text is the main text describing the search target, calibration keywords strongly associated with the search target will be relatively evenly distributed in the text. If there is content unrelated to the search target in the visual clustering region, there may be no calibration keywords strongly associated with the search target in this part of the content. Therefore, for the special case where the calibration keywords are widely spaced, the fuzzy region of the search target is determined, and the target features are removed. Thus, when facing a large number of search interfaces, the irrelevant features on the web page can be effectively reduced, the efficiency and accuracy of obtaining features from the web page can be improved, and the accuracy of the feedback of the language rating model can be ensured.

[0118] So far, the technical solution of the present invention has been described in conjunction with the preferred embodiments shown in the drawings. However, it is easy for those skilled in the art to understand that the protection scope of the present invention is obviously not limited to these specific embodiments. Without departing from the principle of the present invention, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after these changes or substitutions will all fall within the protection scope of the present invention.

Claims

1. A rating feature classification method for a technology enterprise rating model, characterized in that, Including: Set a retrieval target and obtain a number of retrieval interfaces associated with the retrieval target; Perform interface visual analysis on the retrieval interfaces, including identifying a number of target features in the page, clustering the target features based on the relative position relationship between the target features, and determining a number of visual clustering regions; Extract some of the target features in the visual clustering regions, determine the relevance parameter between the target features and the retrieval target, and set classification labels for each of the visual clustering regions; Analyze and extract the target features in each visual clustering region according to the classification labels, including Identifying the calibration keywords strongly associated with the retrieval target in the visual clustering region, determining the position information of the calibration keywords in the visual clustering region, screening the calibration keyword groups for the position of the retrieval target fuzzy region, determining the retrieval target fuzzy region in the visual clustering region, removing the target features in the retrieval target fuzzy region, and retaining the remaining target features; Or, remove all the target features in the visual clustering region; Transmit the retained target features to a preset language rating model; Wherein, the retrieval target is in text form and is set by the user side.

2. The rating feature classification method for the technology enterprise rating model according to claim 1, characterized in that, The process of clustering the target features based on the relative position relationship between the target features and determining a number of visual clustering regions includes Determining the coordinates of a number of target features in the retrieval interface; Determining the relative distance between the target features based on each of the coordinates; Clustering each target feature based on the relative distance to obtain a number of clustering clusters; Determining each edge target feature in the clustering cluster, and connecting the edge target features in sequence to obtain a connection region; Enlarge the connection region by a predetermined multiple to obtain the visual clustering region; Wherein, the relative distance between the target features in each clustering cluster and the nearest target feature is less than a predetermined distance threshold.

3. The rating feature classification method for the technology enterprise rating model according to claim 1, characterized in that The process of extracting some of the target features in the visual clustering region includes Dividing the text composed of the target features in the visual clustering region into a number of text segments; Extracting a number of text segments at a predetermined extraction interval.

4. The rating feature classification method for the technology enterprise rating model according to claim 1, characterized in that The process of determining the relevance parameter between the target features and the retrieval target includes Determining the text segments corresponding to the extracted part of the target features in the visual clustering region; Comparing the text segments with the retrieval target respectively to determine the semantic relevance degree corresponding to each text segment, and determining the average value of the semantic relevance degree as the relevance parameter.

5. The rating feature classification method for the technology enterprise rating model according to claim 1, characterized in that, Setting classification labels for each of the visual clustering regions includes If the relevance parameter is greater than or equal to a predetermined relevance standard threshold, set an association label for the visual clustering region; If the relevance parameter is less than the predetermined relevance standard threshold, set a fuzzy label for the visual clustering region.

6. The rating feature classification method for the technology enterprise rating model according to claim 1, wherein, Analyzing and extracting the target features in each visual clustering region according to the classification labels includes If the visual clustering region is an association label, identify the calibration keywords strongly associated with the retrieval target in the visual clustering region, determine the position information of the calibration keywords in the visual clustering region, screen the calibration keyword groups for the position of the retrieval target fuzzy region, determine the retrieval target fuzzy region in the visual clustering region, remove the target features in the retrieval target fuzzy region, and retain the remaining target features; If the visual clustering region is a fuzzy label, all target features in the visual clustering region are removed.

7. The rating feature classification method for the technology enterprise rating model according to claim 1, wherein The process of screening calibration keyword groups for reference to the fuzzy region spacing includes determining the distance between each calibration keyword and the nearest adjacent calibration keyword in the reference direction based on the position information of the calibration keywords; if there is a distance between a calibration keyword and the nearest adjacent calibration keyword greater than a predetermined fuzzy region spacing threshold; then the calibration keyword and the nearest adjacent calibration keyword are determined as a calibration keyword group; wherein the reference direction is perpendicular to the bottom edge direction of the retrieval interface.

8. The rating feature classification method for the technology enterprise rating model according to claim 1, characterized in that, The process of determining the retrieval target fuzzy region in the visual clustering region includes constructing a reference line parallel to the reference direction at the center of the visual clustering region; constructing a calibration line perpendicular to the reference line on the reference line passing through the position of the keyword in the calibration keyword group. The region enclosed by each calibration line and the edge of the visual clustering region is determined as the retrieval target fuzzy region.

9. The rating feature classification method for the technology enterprise rating model according to claim 1, wherein Obtain a number of retrieval interfaces associated with the retrieval target, wherein if the semantic association degree between the title corresponding to the retrieval interface and the retrieval target is greater than a predetermined association degree standard threshold, it is determined that the retrieval interface is associated with the retrieval target.

10. The rating feature classification method for the technology enterprise rating model according to claim 1, characterized in that, The target features include text features and symbol features.

Images