Assessment index and evidence material matching method, device and equipment

Abstract

The invention discloses an assessment index and evidence material matching method, device and equipment. The method comprises the following steps: obtaining a text representation of an assessment index and an evidence material set corresponding to the assessment index; obtaining a first feature vector based on the text representation of the assessment index through a first extraction branch of the semantic matching model; obtaining a second feature vector corresponding to the first evidence material based on the text representation of the first evidence material in the evidence material set through a second extraction branch of the semantic matching model; through a matching layer of a semantic matching model, based on the first feature vector and a second feature vector corresponding to the first evidence material, calculating a matching probability of the assessment index and the first evidence material; and adjusting parameters of the semantic matching model based on the assessment index and the matching probability corresponding to each evidence material in the evidence material set. And the adjusted semantic matching model is used for matching the assessment index and the evidence material, so that the matching efficiency can be improved.

Description

Technical Field

[0001] The embodiments of the present application relate to the field of computer technology, and in particular to a method, device and equipment for matching assessment indicators with supporting materials. Background Art

[0002] During the development of an enterprise, the superior department will formulate various assessment indicators (such as performance indicators and cultural construction indicators) for the subordinate departments. In order to facilitate the superior department to accept the development results, the subordinate departments need to submit corresponding supporting materials for different assessment indicators.

[0003] At present, assessment indicators and supporting materials are mostly matched manually to ensure that correct supporting materials can be uploaded for different assessment indicators.

[0004] However, the above manual matching method takes a lot of time and has low efficiency when the number of supporting materials and assessment indicators is large. Summary of the invention

[0005] The embodiment of the present application provides a method, device and equipment for matching assessment indicators with supporting materials. The technical solution provided by the embodiment of the present application is as follows:

[0006] According to one aspect of an embodiment of the present application, a method for matching assessment indicators with supporting materials is provided, the method comprising:

[0007] Obtaining a text representation of an assessment indicator and a set of supporting materials corresponding to the assessment indicator, wherein the set of supporting materials includes at least one relevant material and at least one irrelevant material, wherein the relevant material refers to supporting materials that match the assessment indicator, and the irrelevant material refers to supporting materials that do not match the assessment indicator;

[0008] Obtaining a first feature vector based on the text representation of the assessment indicator through a first extraction branch of the semantic matching model, wherein the first feature vector is used to characterize a semantic feature of the text representation of the assessment indicator;

[0009] Obtaining, by the second extraction branch of the semantic matching model, a second feature vector corresponding to the first supporting material based on the text representation of the first supporting material in the supporting material set, wherein the second feature vector is used to characterize the semantic features of the text representation of the supporting material;

[0010] By means of the matching layer of the semantic matching model, based on the first feature vector and the second feature vector corresponding to the first supporting material, a matching probability between the assessment indicator and the first supporting material is calculated;

[0011] Based on the matching probabilities respectively corresponding to the assessment indicators and each piece of supporting material in the supporting material set, the parameters of the semantic matching model are adjusted to obtain an adjusted semantic matching model, and the adjusted semantic matching model is used to match the assessment indicators and supporting materials.

[0012] According to one aspect of an embodiment of the present application, a method for matching assessment indicators with supporting materials is provided, the method comprising:

[0013] Obtain the textual representation of the assessment indicators and at least one piece of supporting material;

[0014] Obtaining a first feature vector based on the text representation of the assessment indicator through a first extraction branch of the semantic matching model, wherein the first feature vector is used to characterize a semantic feature of the text representation of the assessment indicator;

[0015] Obtaining, by means of the second extraction branch of the semantic matching model, second feature vectors corresponding to the respective pieces of the supporting materials based on the text representation of the at least one piece of supporting material, wherein the second feature vectors are used to characterize semantic features of the text representation of the supporting material;

[0016] Calculating the matching probabilities of the assessment indicators and the supporting materials respectively based on the first feature vector and the second feature vectors respectively corresponding to the supporting materials through the matching layer of the semantic matching model;

[0017] Based on the matching probabilities respectively corresponding to the assessment indicator and each piece of the supporting material, a matching material matching the assessment indicator is determined from each piece of the supporting material.

[0018] According to one aspect of an embodiment of the present application, a device for matching assessment indicators with supporting materials is provided, the device comprising:

[0019] An acquisition module, used to acquire a text representation of an assessment indicator and a set of supporting materials corresponding to the assessment indicator, wherein the set of supporting materials includes at least one relevant material and at least one irrelevant material, wherein the relevant material refers to supporting materials that match the assessment indicator, and the irrelevant material refers to supporting materials that do not match the assessment indicator;

[0020] A first extraction module, configured to obtain a first feature vector based on the text representation of the assessment indicator through a first extraction branch of a semantic matching model, wherein the first feature vector is used to characterize a semantic feature of the text representation of the assessment indicator;

[0021] A second extraction module, configured to obtain, through a second extraction branch of the semantic matching model, a second feature vector corresponding to the first supporting material based on the text representation of the first supporting material in the supporting material set, wherein the second feature vector is used to characterize the semantic features of the text representation of the supporting material;

[0022] A matching module, configured to calculate, through a matching layer of the semantic matching model, a matching probability between the assessment indicator and the first supporting material based on the first feature vector and a second feature vector corresponding to the first supporting material;

[0023] The adjustment module is used to adjust the parameters of the semantic matching model based on the matching probabilities corresponding to the assessment indicators and each piece of supporting material in the supporting material set, so as to obtain an adjusted semantic matching model, and the adjusted semantic matching model is used to match the assessment indicators and supporting materials.

[0024] According to one aspect of an embodiment of the present application, a device for matching assessment indicators with supporting materials is provided, the device comprising:

[0025] An acquisition module, used to acquire a text representation of the assessment indicator and a text representation of at least one piece of supporting material;

[0026] A first extraction module, configured to obtain a first feature vector based on the text representation of the assessment indicator through a first extraction branch of a semantic matching model, wherein the first feature vector is used to characterize a semantic feature of the text representation of the assessment indicator;

[0027] A second extraction module, configured to obtain, through a second extraction branch of the semantic matching model, second feature vectors corresponding to each of the supporting materials based on the text representation of the at least one supporting material, wherein the second feature vectors are used to characterize semantic features of the text representation of the supporting material;

[0028] A matching module, configured to calculate the matching probabilities respectively corresponding to the assessment indicator and each piece of the supporting material based on the first feature vector and the second feature vector respectively corresponding to each piece of the supporting material through the matching layer of the semantic matching model;

[0029] The determination module is used to determine the matching materials that match the assessment indicators from the supporting materials based on the matching probabilities corresponding to the assessment indicators and the supporting materials.

[0030] According to one aspect of an embodiment of the present application, a computer device is provided, which includes a processor and a memory, wherein a computer program is stored in the memory, and the computer program is loaded and executed by the processor to implement the above-mentioned method of matching assessment indicators with supporting materials.

[0031] According to one aspect of an embodiment of the present application, a computer-readable storage medium is provided, in which a computer program is stored. The computer program is loaded and executed by a processor to implement the above-mentioned method of matching assessment indicators with supporting materials.

[0032] According to one aspect of an embodiment of the present application, a computer program product is provided, which includes a computer program, and the computer program is stored in a computer-readable storage medium. A processor reads and executes the computer program from the computer-readable storage medium to implement the above-mentioned method of matching assessment indicators with supporting materials.

[0033] The technical solution provided by the embodiments of the present application includes at least the following beneficial effects:

[0034] By obtaining the text representation of the assessment index and the set of supporting materials corresponding to the assessment index, the supporting materials set includes relevant materials matching the assessment index and irrelevant materials not matching the assessment index. Then, through the first extraction branch of the semantic matching model, based on the text representation of the assessment index, a first feature vector is obtained. And, through the second extraction branch of the semantic matching model, based on the text representation of the first supporting material in the supporting material set, a second feature vector corresponding to the first supporting material is obtained, and the first feature vector and the second feature vector are both used to characterize the semantic features of the corresponding text representation. Then, through the matching layer of the semantic matching model, based on the first feature vector and the second feature vector corresponding to the first supporting material, the matching probability of the assessment index and the first supporting material is calculated. Finally, based on the matching probabilities corresponding to each supporting material in the assessment index and the supporting material set, the parameters of the semantic matching model are adjusted to obtain an adjusted semantic matching model. The adjusted semantic matching model is enabled to match the assessment index and the supporting material based on the text representation. Further, using the semantic matching model to match the assessment index and the supporting material can save labor costs and improve matching efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] Figure 1 It is a schematic diagram of an implementation environment of a solution provided by an embodiment of the present application;

[0036] Figure 2 This is a flow chart of a method for matching assessment indicators and supporting materials provided in one embodiment of the present application;

[0037] Figure 3 is a schematic diagram of a semantic matching model structure provided by an embodiment of the present application;

[0038] Figure 4 is a flow chart of a method for matching assessment indicators and supporting materials provided in another embodiment of the present application;

[0039] Figure 5 It is a schematic diagram of a matching scheme of assessment indicators and supporting materials provided in one embodiment of the present application;

[0040] Figure 6 It is a block diagram of a device for matching assessment indicators and supporting materials provided by an embodiment of the present application;

[0041] Figure 7 is a block diagram of a device for matching assessment indicators and supporting materials provided in another embodiment of the present application;

[0042] Figure 8 It is a structural block diagram of a computer device provided by an embodiment of the present application. DETAILED DESCRIPTION

[0043] In order to make the objectives, technical solutions and advantages of the present application clearer, the implementation methods of the present application will be further described in detail below in conjunction with the accompanying drawings.

[0044] Please refer to Figure 1 , which shows a schematic diagram of a solution implementation environment provided by an embodiment of the present application. The solution implementation environment may include a model training device 10 and a model using device 20.

[0045] The model training device 10 may be an electronic device such as a mobile phone, a desktop computer, a tablet computer, a laptop computer, a vehicle-mounted terminal, a server, an intelligent robot, a smart TV, a multimedia playback device, or other electronic devices with strong computing capabilities, which is not limited in this application. The model training device 10 is used to train the semantic matching model 30.

[0046] In the embodiment of the present application, the semantic matching model 30 is a machine learning model. Optionally, the model training device 10 can train the semantic matching model 30 in a machine learning manner so that it has better performance.

[0047] Optionally, the training process of the semantic matching model 30 is as follows (only briefly described here, the specific training process is referred to the following embodiment, which is not repeated here): obtain the text representation of the assessment indicator and the supporting material set corresponding to the assessment indicator, the supporting material set includes at least one relevant material matching the assessment indicator and at least one irrelevant material that does not match the assessment indicator. Through the first extraction branch 40 of the semantic matching model 30, based on the text representation of the assessment indicator, a first feature vector is obtained, and the first feature vector is used to characterize the semantic features of the text representation of the assessment indicator. Through the second extraction branch 50 of the semantic matching model 30, based on the text representation of the first supporting material in the supporting material set, a second feature vector corresponding to the first supporting material is obtained, and the second feature vector is used to characterize the semantic features of the text representation of the supporting material. Through the matching layer 60 of the semantic matching model 30, based on the first feature vector and the second feature vector corresponding to the first supporting material, the matching probability of the assessment indicator and the first supporting material is calculated. Based on the matching probabilities corresponding to the assessment indicator and each supporting material in the supporting material set, the parameters of the semantic matching model are adjusted to obtain the adjusted semantic matching model.

[0048] In some embodiments, the model using device 20 may be an electronic device such as a mobile phone, a desktop computer, a tablet computer, a laptop computer, a vehicle-mounted terminal, a server, an intelligent robot, a smart TV, a multimedia playback device, or other electronic devices with strong computing capabilities, which is not limited in this application. Exemplarily, the adjusted semantic matching model 30 may match the assessment indicators and the supporting materials based on the text representation of the assessment indicators and the text representation of the supporting materials. The specific matching process may be referred to in the following embodiments, which will not be described in detail here.

[0049] The model training device 10 and the model using device 20 can be two independent devices or the same device.

[0050] In the method provided in the embodiment of the present application, the execution subject of each step may be a computer device, which refers to an electronic device with data calculation, processing and storage capabilities. Wherein, when the computer device is a server, the server may be an independent physical server, or a server cluster or distributed system composed of multiple physical servers, or a cloud server providing cloud computing services. The computer device may be Figure 1 The model training device 10 may also be a model using device 20.

[0051] Please refer to Figure 2, which shows a flow chart of a method for matching assessment indicators and supporting materials provided by an embodiment of the present application. The execution subject of each step of the method can be the model training device introduced above. In the following method embodiments, for the sake of ease of description, only the execution subject of each step is introduced as a "computer device". The method may include at least one of the following steps 210 to 250.

[0052] Step 210, obtaining the text representation of the assessment indicators and a set of supporting materials corresponding to the assessment indicators.

[0053] The set of supporting materials includes at least one relevant material and at least one irrelevant material. The relevant material refers to the supporting material that matches the assessment indicators, and the irrelevant material refers to the supporting material that does not match the assessment indicators.

[0054] Assessment indicators are used to indicate the objects to be assessed and evaluated and the corresponding evaluation standards. Assessment indicators may include performance indicators, cultural construction indicators, work attendance indicators, etc. This application does not limit this.

[0055] The textual representation of the assessment indicator refers to the text that represents the content of the assessment indicator. In some embodiments, the content of the assessment indicator may include the assessment points, the assessment object, and the submission requirements of the corresponding supporting materials. The assessment indicator may be represented as a descriptive text. Further, the textual representation of the assessment indicator may include this descriptive text, or may include some keywords in this descriptive text, or may include other texts summarized by the technicians except this descriptive text.

[0056] Supporting materials are used to reflect the results of study, work or construction. Relevant personnel can review the supporting materials to determine whether the party submitting the supporting materials has met the requirements of the corresponding assessment indicators.

[0057] The textual representation of supporting materials refers to the text that represents the content of the supporting materials. In some embodiments, the supporting materials may include multiple documents that reflect the results of study, work or construction. Accordingly, the textual representation of supporting materials may include text such as the numbers, titles, summaries and keywords of the above-mentioned documents.

[0058] Different assessment indicators are matched with different supporting materials. For example, if the assessment indicators include cultural construction indicators and work attendance indicators, and the supporting materials include the "Department Attendance Statistics Table" file and the "Department Cultural Construction Activity Summary" file, then the "Department Attendance Statistics Table" file matches the work attendance indicators, and the "Department Cultural Construction Activity Summary" file matches the cultural construction indicators.

[0059] In some embodiments, the ratio of the relevant materials to the irrelevant materials in the supporting material set is set to 1:1.

[0060] In some embodiments, text containing typos and Chinese abbreviations is constructed in the text representation of the assessment indicators or supporting materials, thereby improving the generalization performance of the semantic matching model.

[0061] Step 220 , obtaining a first feature vector based on the text representation of the assessment indicator through a first extraction branch of the semantic matching model.

[0062] The first feature vector is used to characterize the semantic features of the text representation of the assessment indicator. Semantics refers to the content that can be expressed by the text, and semantic features refer to the features of the content expressed by the text. In other words, when two different text representations express similar content, the two text representations have similar semantic features.

[0063] In some embodiments, please refer to Figure 3 , the first extraction branch 40 includes a first input layer and a first extraction layer, and step 220 includes at least one of the following sub-steps 222 to 224.

[0064] Sub-step 222, through the first input layer, the text representation of the assessment indicator is encoded by word vector encoding and position vector encoding to obtain first encoding information.

[0065] Word vector encoding of text representation refers to segmenting the text representation to obtain a token sequence, and then using a vocabulary (such as the vocabulary of the WordPiece model) to map each token to the corresponding token identifier (tokenids), thereby obtaining the word vector corresponding to each token.

[0066] Position vector encoding refers to encoding the position of each token in the token sequence to obtain the position vector corresponding to each token.

[0067] By adding the word vector and position vector corresponding to the token, the first encoding vector corresponding to the token can be obtained. The first encoding information includes the first encoding vectors of all tokens in the token sequence.

[0068] In some embodiments, the above position vector encoding adopts relative position encoding. Specifically, for the i-th token in the token sequence, the attention score of the j-th token relative to the token can be calculated by the following formula: in, is the attention score, q i is the Query vector corresponding to the i-th token, k jis the key vector of the jth token, is the word vector corresponding to the i-th token, is the word vector corresponding to the jth token, W k，E , W k，R 、u T 、v T are all adjustable weight matrices, R i-j is the relative position code, and ij is greater than 0. Furthermore, after normalizing the attention score, the attention weight can be obtained, and the product of the word vector and the valence matrix (the valence matrix is ​​also an adjustable weight matrix) at each position in the token sequence is weighted and summed according to the attention weight corresponding to each position, and the position vector code of the above i-th token can be obtained.

[0069] Sub-step 224, performing feature extraction on the first encoded information through the first extraction layer to obtain a first feature vector.

[0070] In some embodiments, the first extraction layer may use a Bert (Bidirectional Encoder Representations from Transformer) model for feature extraction.

[0071] The first encoded information obtained by the above method carries the position information of the token sequence, so the first extraction layer can extract more accurate semantic features from the first encoded information.

[0072] Step 230, obtaining a second feature vector corresponding to the first supporting material based on the text representation of the first supporting material in the supporting material set through the second extraction branch of the semantic matching model.

[0073] The second feature vector is used to characterize the semantic features of the text representation of the supporting material.

[0074] In some embodiments, please refer to Figure 3 The second extraction branch 50 includes a second input layer and a second extraction layer, and step 230 includes at least one of the following sub-steps 232 to 234.

[0075] Sub-step 232, through the second input layer, word vector encoding and position vector encoding are performed on the text representation of the first supporting material to obtain second encoding information.

[0076] Sub-step 234, performing feature extraction on the second encoded information through a second extraction layer to obtain a second feature vector.

[0077] The structure of the second extraction branch is the same as that of the first extraction branch. For detailed descriptions of sub-steps 232 to 234 , please refer to sub-steps 222 to 224 , and this application will not go into details therein.

[0078] Step 240, through the matching layer of the semantic matching model, based on the first feature vector and the second feature vector corresponding to the first supporting material, calculate the matching probability between the assessment indicator and the first supporting material.

[0079] In some embodiments, please refer to Figure 3 , the semantic matching model 30 includes a matching layer 60. Step 240 includes at least one of the following sub-steps 242-244.

[0080] Sub-step 242, calculating the cosine similarity between the first feature vector and the second feature vector corresponding to the first supporting material through the matching layer.

[0081] Sub-step 244, determining the cosine similarity as the matching probability between the assessment indicator and the first supporting material.

[0082] Right now Among them, y cri is the first eigenvector, y sup is the second eigenvector corresponding to the first supporting material, R(y cri ,y su p) is the matching probability between the assessment index and the first supporting material, consin(y cri ,y sup ) is the cosine similarity between the first eigenvector and the second eigenvector corresponding to the first supporting material.

[0083] Step 250, based on the assessment indicators and the matching probabilities corresponding to each piece of supporting material in the supporting material set, the parameters of the semantic matching model are adjusted to obtain an adjusted semantic matching model.

[0084] The adjusted semantic matching model is used to match assessment indicators and supporting materials.

[0085] In some embodiments, step 250 includes at least one of the following sub-steps 252 to 256 .

[0086] Sub-step 252, based on the assessment index and the matching probability corresponding to each piece of supporting material in the supporting material set, calculate the cross entropy corresponding to each piece of relevant material.

[0087] In some embodiments, for the first related material in each piece of related material, the corresponding cross entropy may be calculated by the following formula: Among them, y sup1 is the second eigenvector corresponding to the first related material, R(ycri ,y sup1 ) is the matching probability between the assessment index and the first relevant material, P(y sup1 |y cri ) is the cross entropy corresponding to the first related material, Y sup is the set of second eigenvectors corresponding to each piece of supporting material in the supporting material set, y′ sup is the second eigenvector corresponding to any piece of supporting material in the set, R(y cri , y′ sup ) is the matching probability between the assessment index and any of the aforementioned supporting materials, and γ is a hyperparameter.

[0088] Sub-step 254, based on the cross entropy corresponding to each piece of relevant material, calculate the loss of the semantic matching model.

[0089] The above loss is used to reflect the performance of the semantic matching model in matching assessment indicators and supporting materials.

[0090] In some embodiments, the cross entropy loss corresponding to the supporting material set may be calculated by the following formula: in, is the second eigenvector corresponding to the relevant material, The cross entropy corresponding to the relevant material is, It is the product of the cross entropy corresponding to all relevant materials in the supporting material set. This cross entropy loss can be used as the loss of the semantic matching model.

[0091] Sub-step 256, based on the above loss, adjust the parameters of the semantic matching model to obtain an adjusted semantic matching model.

[0092] In some embodiments, with the goal of minimizing the above-mentioned loss, the parameters of the semantic matching model are adjusted to obtain an adjusted semantic matching model.

[0093] The above method adjusts the parameters of the semantic matching model by calculating the cross entropy loss, which can improve the ability of the semantic matching model to judge whether the assessment indicators and supporting materials match.

[0094] The technical solution provided by the embodiment of the present application obtains the text representation of the assessment index and the supporting material set corresponding to the assessment index, and the supporting material set includes relevant materials matching the assessment index and irrelevant materials not matching the assessment index. Then, through the first extraction branch of the semantic matching model, a first feature vector is obtained based on the text representation of the assessment index. And, through the second extraction branch of the semantic matching model, based on the text representation of the first supporting material in the supporting material set, a second feature vector corresponding to the first supporting material is obtained, and the first feature vector and the second feature vector are both used to characterize the semantic features of the corresponding text representation. Then, through the matching layer of the semantic matching model, based on the first feature vector and the second feature vector corresponding to the first supporting material, the matching probability of the assessment index and the first supporting material is calculated. Finally, based on the matching probabilities corresponding to each supporting material in the assessment index and the supporting material set, the parameters of the semantic matching model are adjusted to obtain an adjusted semantic matching model. The adjusted semantic matching model is enabled to match the assessment index and the supporting material based on the text representation. Further, using the semantic matching model to match the assessment index and the supporting material can save labor costs and improve matching efficiency.

[0095] Below, the use process of the above-mentioned semantic matching language model is introduced and explained through an embodiment. The contents involved in the above-mentioned model training process and the contents involved in the use process correspond to each other, and the two are interconnected. If one side is not explained in detail, you can refer to the description on the other side.

[0096] Please refer to Figure 4 , which shows a flow chart of a method for matching assessment indicators and supporting materials provided by another embodiment of the present application. The execution subject of each step of the method can be the model use device introduced above. In the following method embodiments, for the convenience of description, only the execution subject of each step is introduced and explained as a "computer device". The method may include at least one of the following steps 410 to 450.

[0097] Step 410, obtaining a textual representation of the assessment indicator and a textual representation of at least one piece of supporting material.

[0098] Step 420 , obtaining a first feature vector based on the text representation of the assessment indicator through a first extraction branch of the semantic matching model.

[0099] The first feature vector is used to characterize the semantic features of the text representation of the assessment indicator.

[0100] In some embodiments, the first extraction branch includes a first input layer and a first extraction layer.

[0101] Step 420 includes at least one of the following sub-steps 422 to 424 .

[0102] Sub-step 422, through the first input layer, the text representation of the assessment indicator is encoded by word vector encoding and position vector encoding to obtain first encoding information.

[0103] Sub-step 424, performing feature extraction on the first encoded information through the first extraction layer to obtain a first feature vector.

[0104] Step 430, obtaining second feature vectors corresponding to each piece of supporting material based on the text representation of at least one piece of supporting material through the second extraction branch of the semantic matching model.

[0105] The second feature vector is used to characterize the semantic features of the text representation of the supporting material.

[0106] Step 440, through the matching layer of the semantic matching model, based on the first feature vector and the second feature vector corresponding to each piece of supporting material, the matching probability corresponding to the assessment indicator and each piece of supporting material is calculated.

[0107] Step 450, based on the matching probabilities corresponding to the assessment indicators and the supporting materials, determine the matching materials that match the assessment indicators from the supporting materials.

[0108] In some embodiments, step 450 includes at least one of the following sub-steps 452 - 454 .

[0109] Sub-step 452, selecting a first number of supporting materials as candidate materials from each piece of supporting material in descending order of matching probability.

[0110] In some embodiments, the supporting materials are first sorted in descending order of matching probability, and then the first number of supporting materials ranked first are selected as candidate materials.

[0111] The first quantity can be set by the technician as needed, such as 2 pieces, 3 pieces, or 4 pieces, and this application does not limit this.

[0112] Sub-step 454, determining the supporting materials with the matching probability greater than the first threshold among the candidate materials as matching materials.

[0113] That is to say, among the candidate materials, the supporting materials whose matching probability with the assessment index is greater than the first threshold are determined as matching materials.

[0114] The first threshold value can be set by the technician as needed, such as 60%, 75%, or 90%, and this application does not limit this.

[0115] The above method matches the assessment indicators and supporting materials through a semantic matching model (the semantic matching model directly outputs the matching probabilities corresponding to the assessment indicators and each piece of supporting material based on the text representation), which saves labor costs and improves matching efficiency.

[0116] Please refer to Figure 5 , which shows a schematic diagram of a matching scheme of assessment indicators and supporting materials provided by an embodiment of the present application, and the scheme includes at least one of the following steps.

[0117] 1. Input the text representation of the assessment indicator and the text representation of at least one piece of supporting material into the semantic matching model 30.

[0118] 2. The semantic matching model 30 is used to obtain the matching probabilities corresponding to the assessment indicators and each piece of supporting material.

[0119] 3. Sort the above matching probabilities in descending order.

[0120] 4. Select supporting materials whose matching probability with the assessment indicators is greater than the first threshold as matching materials.

[0121] The following is an embodiment of the device of the present application, which can be used to execute the embodiment of the method of the present application. For details not disclosed in the embodiment of the device of the present application, please refer to the embodiment of the method of the present application.

[0122] Please refer to Figure 6 , which shows a block diagram of a device for matching assessment indicators and supporting materials provided by an embodiment of the present application. The device can be a computer device, or can be set in a computer device. The device 600 can include: an acquisition module 610, a first extraction module 620, a second extraction module 630, a matching module 640, and an adjustment module 650.

[0123] Acquisition module 610 is used to obtain the text representation of the assessment indicator and the set of supporting materials corresponding to the assessment indicator, the set of supporting materials includes at least one relevant material and at least one irrelevant material, the relevant material refers to the supporting material that matches the assessment indicator, and the irrelevant material refers to the supporting material that does not match the assessment indicator.

[0124] The first extraction module 620 is used to obtain a first feature vector based on the text representation of the assessment indicator through a first extraction branch of a semantic matching model, where the first feature vector is used to characterize the semantic features of the text representation of the assessment indicator.

[0125] The second extraction module 630 is used to obtain a second feature vector corresponding to the first supporting material in the supporting material set based on the text representation of the first supporting material through the second extraction branch of the semantic matching model, wherein the second feature vector is used to characterize the semantic features of the text representation of the supporting material.

[0126] The matching module 640 is used to calculate the matching probability between the assessment indicator and the first supporting material based on the first feature vector and the second feature vector corresponding to the first supporting material through the matching layer of the semantic matching model.

[0127] The adjustment module 650 is used to adjust the parameters of the semantic matching model based on the matching probabilities corresponding to the assessment indicators and each piece of supporting material in the supporting material set, so as to obtain an adjusted semantic matching model, and the adjusted semantic matching model is used to match the assessment indicators and supporting materials.

[0128] In some embodiments, the first extraction branch includes a first input layer and a first extraction layer. The first extraction module 620 is used to perform word vector encoding and position vector encoding on the text representation of the assessment indicator through the first input layer to obtain first encoding information; and perform feature extraction on the first encoding information through the first extraction layer to obtain the first feature vector.

[0129] In some embodiments, the matching module 640 is used to calculate the cosine similarity between the first feature vector and the second feature vector corresponding to the first supporting material through the matching layer; and determine the cosine similarity as the matching probability between the assessment indicator and the first supporting material.

[0130] In some embodiments, the adjustment module 650 is used to calculate the cross entropy corresponding to each piece of the relevant material based on the matching probability corresponding to each piece of supporting material in the supporting material set and the assessment index; calculate the loss of the semantic matching model based on the cross entropy corresponding to each piece of the relevant material; and adjust the parameters of the semantic matching model based on the loss to obtain the adjusted semantic matching model.

[0131] Please refer to Figure 7 , which shows a block diagram of a device for matching assessment indicators and supporting materials provided by another embodiment of the present application. The device may be a computer device, or may be set in a computer device. The device 700 may include: an acquisition module 710, a first extraction module 720, a second extraction module 730, a matching module 740, and a determination module 750.

[0132] The acquisition module 710 is used to acquire the text representation of the assessment indicator and the text representation of at least one piece of supporting material.

[0133] The first extraction module 720 is used to obtain a first feature vector based on the text representation of the assessment indicator through a first extraction branch of a semantic matching model, where the first feature vector is used to characterize the semantic features of the text representation of the assessment indicator.

[0134] The second extraction module 730 is used to obtain a second feature vector corresponding to each piece of supporting material based on the text representation of the at least one supporting material through the second extraction branch of the semantic matching model, wherein the second feature vector is used to characterize the semantic features of the text representation of the supporting material.

[0135] The matching module 740 is used to calculate the matching probability between the assessment indicator and each piece of the supporting material based on the first feature vector and the second feature vector corresponding to each piece of the supporting material through the matching layer of the semantic matching model.

[0136] The determination module 750 is used to determine the matching materials that match the assessment indicators from the supporting materials based on the matching probabilities corresponding to the assessment indicators and the supporting materials.

[0137] In some embodiments, the first extraction branch includes a first input layer and a first extraction layer, and the first extraction module 720 is used to perform word vector encoding and position vector encoding on the text representation of the assessment indicator through the first input layer to obtain first encoding information; and perform feature extraction on the first encoding information through the first extraction layer to obtain the first feature vector.

[0138] In some embodiments, the determination module 750 is used to select a first number of supporting materials from each piece of the supporting materials as candidate materials in the order of the matching probability from high to low; and determine the supporting materials among the candidate materials whose matching probability is greater than a first threshold as the matching materials.

[0139] It should be noted that the device provided in the above embodiment, when implementing its functions, is only illustrated by the division of the above functional modules. In actual applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the device is divided into different functional modules to complete all or part of the functions described above. In addition, the device and method embodiments provided in the above embodiment belong to the same concept, and the specific implementation process is detailed in the method embodiment, which will not be repeated here.

[0140] Please refer to Figure 8 , which exemplarily shows a structural block diagram of a computer device provided by an embodiment of the present application.

[0141] Typically, the computer device 800 includes a processor 801 and a memory 802 .

[0142] The processor 801 may include one or more processing cores, such as a 4-core processor, an 8-core processor, and the like. The processor 801 may be implemented in at least one hardware form of DSP (Digital Signal Processing), FPGA (Field Programmable Gate Array), and PLA (Programmable Logic Array). The processor 801 may also include a main processor and a coprocessor. The main processor is a processor for processing data in an awake state, also known as a CPU (Central Processing Unit); the coprocessor is a low-power processor for processing data in a standby state. In some embodiments, the processor 801 may be integrated with a GPU (Graphics Processing Unit), which is responsible for rendering and drawing the content to be displayed on the display screen. In some embodiments, the processor 801 may also include an AI processor for processing computing operations related to machine learning.

[0143] The memory 802 may include one or more computer-readable storage media, which may be tangible and non-transitory. The memory 802 may also include high-speed random access memory, and non-volatile memory, such as one or more disk storage devices, flash memory storage devices. In some embodiments, the non-transitory computer-readable storage medium in the memory 802 stores a computer program, which is loaded and executed by the processor 801 to implement the above-mentioned matching method of assessment indicators and supporting materials.

[0144] Those skilled in the art will understand that Figure 8 The structure shown in the figure does not constitute a limitation on the computer device 800, and the computer device 800 may include more or less components than those shown in the figure, or combine some components, or adopt a different arrangement of components.

[0145] In some embodiments, a computer-readable storage medium is also provided, in which a computer program is stored. The computer program is loaded and executed by a processor to implement the above-mentioned method of matching assessment indicators with supporting materials.

[0146] Optionally, the computer readable storage medium may include: ROM (Read-Only Memory), RAM (Random-Access Memory), SSD (Solid State Drives) or optical disks, etc. Among them, the random access memory may include ReRAM (Resistance Random Access Memory) and DRAM (Dynamic Random Access Memory).

[0147] In some embodiments, a computer program product is also provided, which includes a computer program, and the computer program is stored in a computer-readable storage medium. The processor reads and executes the computer program from the computer-readable storage medium to implement the above-mentioned method of matching assessment indicators with supporting materials.

[0148] It should be understood that the "multiple" mentioned in this article refers to two or more. "And / or" describes the association relationship of associated objects, indicating that three relationships may exist. For example, A and / or B can represent: A exists alone, A and B exist at the same time, and B exists alone. The character " / " generally indicates that the objects associated before and after are in an "or" relationship. In addition, the step numbers described in this article only illustrate a possible execution sequence between the steps. In some other embodiments, the above steps may not be executed in the order of the numbers, such as two steps with different numbers are executed at the same time, or two steps with different numbers are executed in the opposite order to the diagram. The embodiments of the present application are not limited to this.

[0149] The above are merely exemplary embodiments of the present application and are not intended to limit the present application. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present application shall be included in the protection scope of the present application.

Claims

1. A method for matching assessment indicators with supporting materials, characterized in that: The method comprises: Obtaining a text representation of an assessment indicator and a set of supporting materials corresponding to the assessment indicator, wherein the set of supporting materials includes at least one relevant material and at least one irrelevant material, wherein the relevant material refers to supporting materials that match the assessment indicator, and the irrelevant material refers to supporting materials that do not match the assessment indicator; Obtaining a first feature vector based on the text representation of the assessment indicator through a first extraction branch of the semantic matching model, wherein the first feature vector is used to characterize a semantic feature of the text representation of the assessment indicator; Obtaining, by the second extraction branch of the semantic matching model, a second feature vector corresponding to the first supporting material based on the text representation of the first supporting material in the supporting material set, wherein the second feature vector is used to characterize the semantic features of the text representation of the supporting material; By means of the matching layer of the semantic matching model, based on the first feature vector and the second feature vector corresponding to the first supporting material, a matching probability between the assessment indicator and the first supporting material is calculated; Based on the matching probabilities respectively corresponding to the assessment indicators and each piece of supporting material in the supporting material set, the parameters of the semantic matching model are adjusted to obtain an adjusted semantic matching model, and the adjusted semantic matching model is used to match the assessment indicators and supporting materials.

2. The method according to claim 1, characterized in that The first extraction branch includes a first input layer and a first extraction layer; The first extraction branch of the semantic matching model is used to obtain a first feature vector based on the text representation of the assessment indicator, including: Through the first input layer, word vector encoding and position vector encoding are performed on the text representation of the assessment indicator to obtain first encoding information; The first extraction layer performs feature extraction on the first coded information to obtain the first feature vector.

3. The method according to claim 1, characterized in that The calculating, through the matching layer of the semantic matching model, the matching probability between the assessment indicator and the first supporting material based on the first feature vector and the second feature vector corresponding to the first supporting material, includes: Calculating, by the matching layer, the cosine similarity between the first feature vector and the second feature vector corresponding to the first supporting material; The cosine similarity is determined as the matching probability between the assessment indicator and the first supporting material.

4. The method according to claim 1, characterized in that The adjusting of the parameters of the semantic matching model based on the assessment index and the matching probabilities corresponding to each piece of supporting material in the supporting material set to obtain the adjusted semantic matching model includes: Based on the assessment index and the matching probabilities corresponding to each piece of supporting material in the supporting material set, the cross entropy corresponding to each piece of the relevant material is calculated; Calculating the loss of the semantic matching model based on the cross entropy corresponding to each piece of the relevant material; Based on the loss, the parameters of the semantic matching model are adjusted to obtain the adjusted semantic matching model.

5. A method for matching assessment indicators with supporting materials, characterized in that: The method comprises: Obtain the textual representation of the assessment indicators and at least one piece of supporting material; Obtaining a first feature vector based on the text representation of the assessment indicator through a first extraction branch of the semantic matching model, wherein the first feature vector is used to characterize a semantic feature of the text representation of the assessment indicator; Obtaining, by means of the second extraction branch of the semantic matching model, second feature vectors corresponding to the respective pieces of the supporting materials based on the text representation of the at least one piece of supporting material, wherein the second feature vectors are used to characterize semantic features of the text representation of the supporting material; Calculating the matching probabilities of the assessment indicators and the supporting materials respectively based on the first feature vector and the second feature vectors respectively corresponding to the supporting materials through the matching layer of the semantic matching model; Based on the matching probabilities respectively corresponding to the assessment indicator and each piece of the supporting material, a matching material matching the assessment indicator is determined from each piece of the supporting material.

6. The method according to claim 5, characterized in that The first extraction branch includes a first input layer and a first extraction layer; The first extraction branch of the semantic matching model is used to obtain a first feature vector based on the text representation of the assessment indicator, including: Through the first input layer, word vector encoding and position vector encoding are performed on the text representation of the assessment indicator to obtain first encoding information; The first extraction layer performs feature extraction on the first coded information to obtain the first feature vector.

7. The method according to claim 5, characterized in that The determining, based on the matching probabilities respectively corresponding to the assessment index and each piece of the supporting material, from each piece of the supporting material, a matching material matching the assessment index comprises: Selecting a first number of supporting materials from each of the supporting materials as candidate materials in descending order of the matching probabilities; The supporting material of the candidate materials whose matching probability is greater than a first threshold is determined as the matching material.

8. A device for matching assessment indicators and supporting materials, characterized in that: The device comprises: An acquisition module, used to acquire a text representation of an assessment indicator and a set of supporting materials corresponding to the assessment indicator, wherein the set of supporting materials includes at least one relevant material and at least one irrelevant material, wherein the relevant material refers to supporting materials that match the assessment indicator, and the irrelevant material refers to supporting materials that do not match the assessment indicator; A first extraction module, configured to obtain a first feature vector based on the text representation of the assessment indicator through a first extraction branch of a semantic matching model, wherein the first feature vector is used to characterize a semantic feature of the text representation of the assessment indicator; A second extraction module, configured to obtain, through a second extraction branch of the semantic matching model, a second feature vector corresponding to the first supporting material based on the text representation of the first supporting material in the supporting material set, wherein the second feature vector is used to characterize the semantic features of the text representation of the supporting material; A matching module, configured to calculate, through a matching layer of the semantic matching model, a matching probability between the assessment indicator and the first supporting material based on the first feature vector and a second feature vector corresponding to the first supporting material; The adjustment module is used to adjust the parameters of the semantic matching model based on the matching probabilities corresponding to the assessment indicators and each piece of supporting material in the supporting material set, so as to obtain an adjusted semantic matching model, and the adjusted semantic matching model is used to match the assessment indicators and supporting materials.

9. A device for matching assessment indicators and supporting materials, characterized in that: The device comprises: An acquisition module, used to acquire a text representation of the assessment indicator and a text representation of at least one piece of supporting material; A first extraction module, configured to obtain a first feature vector based on the text representation of the assessment indicator through a first extraction branch of a semantic matching model, wherein the first feature vector is used to characterize a semantic feature of the text representation of the assessment indicator; A second extraction module, configured to obtain, through a second extraction branch of the semantic matching model, second feature vectors corresponding to each of the supporting materials based on the text representation of the at least one supporting material, wherein the second feature vectors are used to characterize semantic features of the text representation of the supporting material; A matching module, configured to calculate the matching probabilities respectively corresponding to the assessment indicator and each piece of the supporting material based on the first feature vector and the second feature vector respectively corresponding to each piece of the supporting material through the matching layer of the semantic matching model; The determination module is used to determine the matching materials that match the assessment indicators from the supporting materials based on the matching probabilities corresponding to the assessment indicators and the supporting materials.

10. A computer device, characterized in that: The computer device includes a processor and a memory, wherein a computer program is stored in the memory, and the computer program is loaded and executed by the processor to implement the method according to any one of claims 1 to 4, or to implement the method according to any one of claims 5 to 7.

Images