CN114282513A - Matching method, system, intelligent terminal and storage medium for text semantic similarity - Google Patents

Abstract

The present application relates to natural language processing technology in the field of artificial intelligence, in particular to a method, system, intelligent terminal and storage medium for matching text semantic similarity, wherein the method includes acquiring historical data as a training sample set, and the training sample set includes True sample, positive sample and negative sample; calculate the cosine similarity between the true sample and the positive sample and the cosine similarity between the true sample and the negative sample, and train the preset text based on the calculation result A semantic similarity matching model; deploying the text semantic similarity matching model to an online platform; matching standard questions based on the text semantic similarity matching model and feeding back to the online platform. The present application can improve the problem of low accuracy of customer service problem matching.

Description

Matching method, system, intelligent terminal and storage medium for text semantic similarity

technical field

The present application relates to natural language processing technology in the field of artificial intelligence, in particular to a method, system, intelligent terminal and storage medium for matching text semantic similarity.

Background technique

With the rapid development of computer Internet, text similarity calculation has been widely used in many fields, especially in the current customer service problem matching scenario; the process of customer service problem matching scenario is as follows: customer service responds to the questions raised by users, by judging the problem Text similarity, retrieve similar questions in the database, and feed back the retrieved questions to the user. In the current customer service problem matching scenario, the method for judging the text similarity is mainly to evaluate the text similarity based on the word frequency, that is, to count the number of occurrences of each word in the two texts, to construct a text vector according to the number of occurrences of the word, and then to calculate the two Cosine similarity between text vectors, reflecting the similarity between two texts.

In the process of realizing this application, the inventor found that the above technology has at least the following problems: in the current customer service problem matching scenario, the evaluation of text similarity based on word frequency is separated from the semantic changes brought by the language environment, ignoring the user's language habits, As a result, it is easy to affect the judgment of text similarity, resulting in a low accuracy rate of customer service problem matching.

SUMMARY OF THE INVENTION

In order to improve the problem of low accuracy of customer service question matching, the present application provides a text semantic similarity matching method, system, intelligent terminal and storage medium.

In the first aspect, the present application provides a method for matching text semantic similarity, which adopts the following technical solutions:

A text semantic similarity matching method, comprising the following steps:

Obtaining historical data as a training sample set, the training sample set includes true samples, positive samples and negative samples;

Calculate the cosine similarity between the true sample and the positive sample and the cosine similarity between the true sample and the negative sample, and train a preset text semantic similarity matching model based on the calculation result;

deploying the text semantic similarity matching model to an online platform;

Standard questions are matched based on the text semantic similarity matching model and fed back to the online platform.

By adopting the above technical solution, the historical data is obtained as a training sample set, and the training sample set includes real samples, positive samples and negative samples; and then based on the cosine similarity between the real samples and the positive samples and the cosine similarity between the real samples and the negative samples, the text is analyzed. The semantic similarity matching model is trained. After training, the text semantic similarity matching model is deployed to the online platform and the standard questions are fed back to the online platform. By training the text semantic similarity matching model, the problems and problems actually input by users are improved. The similarity of the standard questions fed back to the user, thereby improving the accuracy of customer service question matching.

In a specific implementation, the real samples include questions that are actually input by the user online; the positive samples include the standard questions selected by the user and the standard questions configured by customer service for the real input of the user; the Negative samples include said standard questions not selected by the user;

By adopting the above technical solution, sufficient training samples are constructed and the samples are divided into detail, which is convenient for model training and improves the accuracy of customer service problem matching.

In a specific implementation, the calculating the cosine similarity between the true sample and the positive sample and the cosine similarity between the true sample and the negative sample, and train a preset based on the calculation result The text semantic similarity matching models include:

Calculate the cosine similarity between the true sample and the positive sample and the cosine similarity between the true sample and the negative sample, and the cosine similarity calculation formula is as follows:

C0 = Cosine(T,P);

C1 = Cosine(T,N1);

Ck = Cosine(T,Nk);

where T represents true samples, P represents positive samples, N represents negative samples, and k represents the number of negative samples;

Constrain the cosine similarity between the true sample and the positive sample to be greater than or equal to the cosine similarity between the true sample and the negative sample, and the constraint formula is as follows:

C0 = Max(C0,C1,…,Ck).

By adopting the above technical solution, since sometimes the cosine similarity between the real sample and the negative sample is greater than the cosine similarity between the real sample and the positive sample when it is separated from the semantic environment, it is necessary in the training process of the text semantic similarity matching model. Always satisfy the cosine similarity between true samples and positive samples is greater than or equal to the cosine similarity between true samples and negative samples.

In a specific implementation, a Softmax function is selected to apply the constraint formula to the cosine similarity calculation formula to obtain Softmax (C0):

Softmax(C0) = Max(Softmax(C0), Softmax(C1),…, Softmax(Ck));

The error between the real problem input by the user and the standard problem selected by the user is defined as Loss, and in the calculation process of Loss, the cosine similarity between the real sample and the positive sample is constrained to be greater than or equal to the real sample. The cosine similarity with the negative sample, Loss is calculated as follows:

Loss = - log(Softmax(C0)).

By adopting the above technical solution, Loss can more vividly represent that the cosine similarity between true samples and positive samples is higher than the cosine similarity between true samples and negative samples. When the text semantic similarity matching model's Loss is smaller, it is considered that the text semantics are similar The more accurate the prediction result of the degree matching model is, the loss can reach the minimum value by training the text semantic similarity matching model, so as to achieve the final purpose of model training.

In a specific implementation, supervised training is performed on the text semantic similarity matching model based on labeled data, where the labeled data includes the standard questions that are actually clicked by the user and questions that are actually input by the user.

By adopting the above technical solution, the text semantic similarity matching model is supervised and trained by labeling data, so that the model can have the ability to predict and classify unknown data.

In a specific implementation, the labeled data is randomly selected as the negative sample relative to the true sample and the positive sample.

By adopting the above technical solution, the labeled data is randomly selected as the negative sample relative to the true sample and the positive sample, and the labeled data is randomly selected as the negative sample, which improves the similarity between the negative sample and the true sample. The similarity is always greater than or equal to the similarity between the real sample and the negative sample, so that the similarity between the positive sample and the real sample is further improved, thereby enhancing the training effect of the text semantic similarity model.

In a specific implementation, the text semantic similarity matching model includes the true sample and the positive sample calculation module and the true sample and the negative sample calculation module;

The deploying the text semantic similarity matching model to the online platform further includes:

The text semantic similarity matching model is cut and the true sample and the positive sample calculation module are retained.

By adopting the above technical solution, the text semantic similarity matching model is composed of a calculation module for real samples and positive samples and a calculation module for real samples and negative samples. The calculation module is convenient to directly call the calculation module of the real sample and the positive sample to match the text similarity in the positive sample set, which can effectively shorten the text similarity matching time and retrieval time, and improve the matching efficiency.

In the second aspect, the present application provides a text semantic similarity matching system, which adopts the following technical solutions:

A text semantic similarity matching system, including:

a data acquisition module for acquiring historical data as a training sample set, the training sample set including true samples, positive samples and negative samples;

The model training module is used to calculate the cosine similarity between the true sample and the positive sample and the cosine similarity between the true sample and the negative sample, and train the preset text semantic similarity based on the calculation result matching model;

a model deployment module for deploying the text semantic similarity matching model to an online platform;

A data feedback module, configured to match standard questions based on the text semantic similarity matching model and feed back to the online platform.

By adopting the above technical solution, the historical data is obtained as a training sample set, and the training sample set includes real samples, positive samples and negative samples; and then based on the cosine similarity between the real samples and the positive samples and the cosine similarity between the real samples and the negative samples, the text is analyzed. The semantic similarity matching model is trained. After training, the text semantic similarity matching model is deployed to the online platform and the standard questions are fed back to the online platform. By training the text semantic similarity matching model, the problems and problems actually input by users are improved. The similarity of the standard questions fed back to the user, thereby improving the accuracy of customer service question matching.

In the third aspect, the application provides an intelligent terminal, which adopts the following technical solutions:

An intelligent terminal, the intelligent terminal includes a processor and a memory, the memory stores at least one instruction, at least a section of program, code set or instruction set, the at least one instruction, the at least section of program, the code The set or instruction set is loaded and executed by the processor to implement a method for matching text semantic similarity according to any one of the first aspect.

By adopting the above technical solution, the processor in the intelligent terminal can implement the above-mentioned method for matching the semantic similarity of text according to the relevant computer program stored in the memory, thereby improving the accuracy of the semantic similarity of the text, thereby improving the accuracy of customer service problem matching. Accuracy.

In a fourth aspect, the present application provides a computer-readable storage medium, which adopts the following technical solutions:

A computer-readable storage medium having stored therein at least one instruction, at least one piece of program, code set or instruction set, said at least one instruction, said at least one piece of program, said code set or instruction set processed by The processor is loaded and executed to implement a text semantic similarity matching method according to any one of the first aspects.

By adopting the above technical solution, a corresponding program can be stored, thereby improving the accuracy of text semantic similarity, thereby improving the accuracy of customer service problem matching.

To sum up, the present application includes at least one of the following beneficial technical effects:

1. Obtain historical data as a training sample set. The training sample set includes real samples, positive samples and negative samples; and then match the text semantic similarity based on the cosine similarity between the real sample and the positive sample and the cosine similarity between the real sample and the negative sample The model is trained. After training, the text semantic similarity matching model is deployed to the online platform and the standard questions are fed back to the online platform. By training the text semantic similarity matching model, the problems actually input by users and the feedback to users can be improved. The similarity of standard questions, thereby improving the accuracy of customer service question matching;

2. Randomly extract the labeled data as the negative sample relative to the true sample and the positive sample. By randomly extracting the labeled data as the negative sample, the similarity between the negative sample and the true sample is improved. Since the similarity between the true sample and the positive sample is always It is greater than or equal to the similarity between the real sample and the negative sample, so that the similarity between the positive sample and the real sample is further improved, thereby enhancing the training effect of the text semantic similarity model;

3. The text semantic similarity matching model is composed of the calculation module of the real sample and the positive sample and the calculation module of the real sample and the negative sample. By removing the calculation module of the real sample and the negative sample in the text semantic similarity matching model, it is convenient to Directly calling the calculation module of the real sample and the positive sample to match the text similarity in the positive sample set can effectively shorten the text similarity matching time and retrieval time, and improve the matching efficiency.

Description of drawings

FIG. 1 is a schematic flowchart of a method for matching text semantic similarity in an embodiment of the present application.

FIG. 2 is a structural block diagram of a text semantic similarity matching system in an embodiment of the present application.

FIG. 3 is a schematic flowchart of a method for matching text semantic similarity in an embodiment of the present application.

Reference numeral description: 100, a data acquisition module; 200, a model training module; 300, a model deployment module; 400, a data feedback module.

Detailed ways

The present application will be further described in detail below with reference to the accompanying drawings.

The embodiment of the present application discloses a text semantic similarity matching method. The method can be applied to an intelligent terminal, with the intelligent terminal as the execution subject, and is used to realize in the matching scenario of customer service problems, according to the actual input of the user online. It extracts the text semantic features in the actual question and retrieves the similar standard questions in the standard question database, judges the most similar standard questions by the text semantic similarity, and retrieves some standard questions with the highest similarity and feeds them back to the user for information. The user makes a selection. Among them, the text semantic similarity refers to extracting high-dimensional semantic features of the text based on the text words, and then calculating the similarity, so as to measure the similarity between different texts.

Referring to Fig. 1, the matching method of text semantic similarity includes the following steps:

S101. Obtain historical data as a training sample set, where the training sample set includes true samples, positive samples and negative samples.

In the implementation, historical data should be obtained first as a training sample set. The training sample set includes real samples, positive samples and negative samples; the real samples are the real input problems of the user's online platform, and the positive samples are the standard problems selected by the user online or The standard questions configured by the customer service staff according to the real input of the user, and the negative samples are the standard questions that the user did not select among the standard questions fed back by the online platform.

S102: Calculate the cosine similarity between the real sample and the positive sample and the cosine similarity between the real sample and the negative sample, and train a preset text semantic similarity matching model based on the calculation result.

In the implementation, high-dimensional semantic features are first extracted from real samples, positive samples and negative samples through a preset text semantic similarity matching model, and Albert is used as the high-dimensional semantic feature extraction of the training sample set in the text semantic similarity matching model. The basic structure of Albert is a deep pre-training model that extracts text features. Compared with other commonly used training models, Albert uses parameter reduction technology to reduce memory consumption and improve the training speed of the model.

In the implementation, the model training is carried out according to the ratio of positive samples: negative samples is 1:4. After extracting the semantic features of real samples, positive samples and negative samples through the text semantic similarity matching model, the cosine similarity between the real samples and the positive samples and the cosine similarity between the real samples and the negative samples are calculated respectively by the cosine similarity formula. Cosine similarity is a measure of the similarity between texts. The closer the cosine similarity is to 1, the higher the similarity between the two is, and the closer to 0, the more independent the two are.

Since the cosine similarity between the real sample and the negative sample is sometimes greater than the cosine similarity between the real sample and the positive sample when it is separated from the semantic environment, it is necessary to always satisfy the real sample and the positive sample during the training process of the text semantic similarity matching model. The cosine similarity of the sample is greater than or equal to the cosine similarity of the true sample and the negative sample.

In order to more vividly indicate that the cosine similarity between the real sample and the positive sample is higher than the cosine similarity between the real sample and the negative sample, the error between the real problem input by the user and the standard problem selected by the user in deep learning is defined as Loss, when The smaller the Loss of the text semantic similarity matching model is, the more accurate the result predicted by the text semantic similarity matching model is considered, and the loss can reach the minimum value by training the text semantic similarity matching model. Among them, when Loss is calculated, the cosine similarity between the true sample and the positive sample is always greater than or equal to the cosine similarity between the true sample and the negative sample.

Specifically, the calculation process of Loss is to first calculate the cosine similarity between the true sample (T) and the positive sample (P) and the cosine similarity between the true sample (T) and the negative sample (N), where k is the number of negative samples (N), and the calculation formula of cosine similarity is as follows:

C0 = Cosine(T,P);

C1 = Cosine(T,N1);

Ck = Cosine(T,Nk);

In order to ensure that the similarity between the true sample and the positive sample is greater than or equal to the similarity between the true sample and the negative sample, the calculation formula of the cosine similarity is constrained. The constraint formula is as follows:

C0 = Max(C0,C1,…,Ck);

The C0 obtained by the constraint is the largest value in the data set composed of the calculated C0, C1, ..., Ck, where C0 in the constraint formula represents the cosine similarity between the true sample and the positive sample, so as to ensure that the true sample and the positive sample are the same. The cosine similarity between positive samples is always greater than the cosine similarity between true samples and negative samples. For example, the maximum value of C1,...,Ck is 0.8, the value of C0 is 0.6, and C0 will be assigned the maximum value of 0.8 by the constraint formula, so as to satisfy the total cosine similarity between the true sample and the positive sample. is greater than or equal to the cosine similarity between true samples and negative samples.

Specifically, since the cosine similarity value range is 0-1, the Softmax function that is also a function and the value range is also 0-1 is selected to apply the constraint formula to the cosine similarity calculation formula of C0, and obtain Softmax (C0), replace the previously calculated C0 with Softmax (C0) as the cosine similarity between the true sample (T) and the positive sample (P). The calculation formula of Softmax (C0) is as follows:

Softmax(C0) = Max(Softmax(C0), Softmax(C1),…, Softmax(Ck));

Since the value of Softmax is at most 1, it is finally necessary to let Softmax (C0) approach 1 wirelessly, so that the cosine similarity between the true sample (T) and the positive sample (P) is the maximum value and is always greater than Equal to the cosine similarity between true samples (T) and negative samples (N).

Since the ultimate purpose of model training is to make Loss reach the minimum value, at this time, the cosine similarity between the true sample (T) and the positive sample (P), that is, Softmax (C0), is the maximum value, so the calculation formula of Loss is as follows:

Loss = - log(Softmax(C0));

In one embodiment, the process of training the text semantic similarity matching model based on the training sample set is supervised training, and the supervised training needs to rely on labeled data, and the labeled data includes the standard question of the user's actual click and the question of the user's actual input . For example, after the user enters the question "How to refund the plane ticket" on the smart terminal, the smart terminal feeds back five standard questions with the highest similarity for the user to choose. The standard questions and real user input questions are annotated and used as a set of labeled data. The supervised training of text semantic similarity matching model by labeling data enables the model to have the ability to predict and classify unknown data.

In one embodiment, in order to improve the training effect of the text semantic similarity model and increase the accuracy of customer service question matching, the labeled data can be randomly selected as the negative sample relative to the true sample and the positive sample, and the labeled data can be randomly selected as the negative sample. The similarity between the negative samples and the real samples is improved. Since the similarity between the real samples and the positive samples is always greater than or equal to the similarity between the real samples and the negative samples, the similarity between the positive samples and the real samples is further improved. to enhance the training effect of the text semantic similarity model.

S103. Deploy the text semantic similarity matching model to an online platform.

Specifically, the text semantic similarity matching model includes a real sample and a positive sample calculation module and a real sample and a negative sample calculation module, and the real sample and positive sample calculation module is specifically the calculated cosine similarity between the real sample and the positive sample The numerical set, the true sample and the negative sample calculation module is specifically the numerical set of the calculated cosine similarity between the true sample and the negative sample. In the implementation, the real sample and negative sample calculation module in the text semantic similarity matching model is removed, and the real sample and positive sample calculation module is retained, which is convenient to directly call the real sample and positive sample calculation module to calculate the text similarity in the positive sample set. It can effectively shorten the text similarity matching time and retrieval time, and improve the matching efficiency.

S104 , matching the standard question based on the text semantic similarity matching model and feeding it back to the online platform.

Specifically, after the user enters a question on the online platform, the text semantic similarity matching model first extracts the semantic features that can be input by the user, and then extracts the semantic features of the positive samples from the positive sample set and matches with the semantic features of the user's question, and finally Five standard questions with the highest similarity are matched and fed back to the user for selection.

Figure 3 shows a schematic flowchart of the matching method of text semantic similarity. The text semantic similarity model first obtains true samples, positive samples and negative samples as training sample sets, and then the text semantic similarity matching model uses Albert as the basic structure to extract High-dimensional semantic features of real samples, positive samples and negative samples, and then calculate the cosine similarity between real samples and positive samples and the cosine similarity between real samples and negative samples, and constrain the cosine similarity between real samples and positive samples to be greater than or equal to The cosine similarity between true samples and negative samples is used to conduct supervised training of the text semantic similarity matching model. After the model training is completed, the calculated values between the true samples, positive samples and negative samples will be back-propagated and stored in the model. middle. When applying the text semantic similarity matching model to the online platform, only the real sample and positive sample calculation modules are retained and deployed to the online platform. After the user enters a question on the online platform, extract the textual semantic features of the question input by the user and automatically search for the most similar positive sample semantic features in the positive sample set, obtain the five most similar positive samples, namely five standard questions, and feed them back to the user. For its selection, in order to complete the text semantic similarity matching process.

The embodiment of the present application also discloses a text semantic similarity matching system. Referring to Figure 2, the matching system for text semantic similarity includes:

The data acquisition module 100 is used for acquiring historical data as a training sample set. The training sample set includes real samples, positive samples and negative samples; the real samples are the real input problems of the user on the online platform, and the positive samples are the standard problems selected by the user online. Or the standard questions configured by the customer service staff for the real input of the user. The negative sample is the standard question that the user did not select among the standard questions fed back by the online platform.

The model training module 200 is used to calculate the cosine similarity between the real sample and the positive sample and the cosine similarity between the real sample and the negative sample, and train a preset text semantic similarity matching model based on the calculation result; The cosine similarity between positive samples and the cosine similarity between true samples and negative samples, and the preset text semantic similarity matching model trained based on the calculation results includes:

Calculate the cosine similarity between the true sample and the positive sample and the cosine similarity between the true sample and the negative sample respectively. The calculation formula of the cosine similarity is as follows:

C0 = Cosine(T,P);

C1 = Cosine(T,N1);

Ck = Cosine(T,Nk);

where T represents true samples, P represents positive samples, N represents negative samples, and k represents the number of negative samples;

Constrain the cosine similarity between true samples and positive samples to be greater than or equal to the cosine similarity between true samples and negative samples. The constraint formula is as follows:

C0 = Max(C0,C1,…,Ck).

Select the Softmax function to apply the constraint formula to the cosine similarity calculation formula to obtain Softmax (C0):

Softmax(C0) = Max(Softmax(C0), Softmax(C1),…, Softmax(Ck));

The error between the real problem input by the user and the standard problem selected by the user is defined as Loss. In the calculation process of Loss, the cosine similarity between the real sample and the positive sample is always greater than or equal to the cosine similarity between the real sample and the negative sample. The formula for calculating Loss is as follows:

Loss = - log(Softmax(C0)).

A model deployment module 300, configured to deploy the text semantic similarity matching model to an online platform;

The data feedback module 400 is used for matching standard questions based on the text semantic similarity matching model and feeding back to the online platform.

Optionally, before the model deployment module 300 includes:

The model cutting module is used to cut the text semantic similarity matching model and retain the real sample and positive sample calculation module.

Optionally, the text semantic similarity matching system further includes:

The supervised training module is used for supervised training of the text semantic similarity matching model based on the labeled data. Annotated data includes standard questions of real user clicks and real input questions of users.

Optionally, the supervised training module includes:

The data enhancement sub-module is used to randomly extract the labeled data as a negative sample relative to the true sample and the positive sample.

The embodiment of the present application further discloses an intelligent terminal, including a memory and a processor, the memory stores a computer program, and when the computer program is executed by the processor, the processor executes the steps of the above text semantic similarity matching method. The steps of a text semantic similarity matching method here may be the steps in the above-mentioned text semantic similarity matching method.

The embodiment of the present application further discloses a computer-readable storage medium, which includes various steps in the flow of the above-mentioned matching method for text semantic similarity when it can be loaded and executed by a processor.

The computer-readable storage medium includes, for example: U disk, removable hard disk, read-only memory (Read-Only Memory, ROM), random access memory (Random Access Memory, RAM), magnetic disk or optical disk and other various storage media that can store program codes. medium.

Those skilled in the art can clearly understand that, for the convenience and simplification of the description, only the division of the above-mentioned functional modules is used for illustration. The internal structure of the device is divided into different functional modules to complete all or part of the functions described above. For the specific working processes of the systems, devices and units described above, reference may be made to the corresponding processes in the foregoing method embodiments, which will not be repeated here. .

The above embodiments are only used to introduce the technical solutions of the present application in detail, but the descriptions of the above embodiments are only used to help understand the method and the core idea of the present application, and should not be construed as a limitation on the present application. Those skilled in the art can easily think of changes or substitutions within the technical scope disclosed in the present application, which should all be covered by the protection scope of the present application.

Claims (10)

1. A matching method of text semantic similarity is characterized by comprising the following steps:

acquiring historical data as a training sample set, wherein the training sample set comprises a true sample, a positive sample and a negative sample;

calculating cosine similarity between the true sample and the positive sample and cosine similarity between the true sample and the negative sample, and training a preset text semantic similarity matching model based on a calculation result;

deploying the text semantic similarity matching model to an online platform;

matching a standard problem based on the text semantic similarity matching model and feeding back the standard problem to the online platform.

2. The matching method of text semantic similarity according to claim 1, characterized in that: the true sample comprises a question of true input on the user line; the positive sample comprises the standard question selected by the user and the standard question of the customer service real input configuration aiming at the user; the negative examples include the standard questions that were not selected by the user.

3. The matching method of text semantic similarity according to claim 2, characterized in that: the calculating the cosine similarity between the true sample and the positive sample and the cosine similarity between the true sample and the negative sample, and the training a preset text semantic similarity matching model based on the calculation result comprises:

respectively calculating cosine similarity between the true sample and the positive sample and cosine similarity between the true sample and the negative sample, wherein a cosine similarity calculation formula is as follows:

C0 = Cosine(T,P)；

C1 = Cosine(T,N1)；

Ck = Cosine(T,Nk)；

wherein T represents a true sample, P represents a positive sample, N represents a negative sample, and k represents the number of negative samples;

and constraining the cosine similarity between the true sample and the positive sample to be more than or equal to the cosine similarity between the true sample and the negative sample, wherein a constraint formula is as follows:

C0 = Max(C0,C1,…,Ck)。

4. the matching method of text semantic similarity according to claim 3, characterized in that: selecting a Softmax function, and applying the constraint formula to the cosine similarity calculation formula to obtain Softmax (C0):

Softmax(C0) = Max(Softmax(C0), Softmax(C1),…, Softmax(Ck))；

defining the error between a real problem input by a user and the standard problem selected by the user as Loss, and constraining the cosine similarity between the true sample and the positive sample to be always more than or equal to the cosine similarity between the true sample and the negative sample in the calculation process of the Loss, wherein the calculation formula of the Loss is as follows:

Loss = - log(Softmax(C0))。

5. the matching method of text semantic similarity according to claim 3, characterized in that: and carrying out supervised training on the text semantic similarity matching model based on labeling data, wherein the labeling data comprise the standard problem really clicked by a user and the problem really input by the user.

6. The matching method of text semantic similarity according to claim 5, characterized in that: randomly extracting the annotation data as the negative sample of the true sample as opposed to the positive sample.

7. The matching method of text semantic similarity according to claim 6, characterized in that:

the text semantic similarity matching model comprises a true sample and positive sample calculation module and a true sample and negative sample calculation module;

before the deploying the text semantic similarity matching model to the online platform, the method further comprises:

and cutting the text semantic similarity matching model and reserving the true sample and the positive sample calculation module.

8. A system for analyzing semantic similarity of texts is characterized by comprising:

a data acquisition module (100) for acquiring historical data as a training sample set, the training sample set comprising true samples, positive samples and negative samples;

the model training module (200) is used for calculating cosine similarity between the true sample and the positive sample and cosine similarity between the true sample and the negative sample, and training a preset text semantic similarity matching model based on a calculation result;

a model deployment module (300) for deploying the text semantic similarity matching model to an online platform;

and the data feedback module (400) is used for matching a standard problem based on the text semantic similarity matching model and feeding back the standard problem to the online platform.

9. An intelligent terminal, characterized in that the intelligent terminal comprises a processor and a memory, wherein the memory stores at least one instruction, at least one program, a set of codes or a set of instructions, and the at least one instruction, the at least one program, the set of codes or the set of instructions is loaded and executed by the processor to realize a matching method for semantic similarity of texts according to any one of claims 1 to 7.

10. A computer-readable storage medium, having stored thereon at least one instruction, at least one program, a set of codes, or a set of instructions, which is loaded and executed by a processor to implement a method of matching semantic similarity of texts as claimed in any one of claims 1 to 7.

Images