Semi-open question type scoring method in voice evaluation scene, storage medium and equipment

Abstract

The invention discloses a semi-open question type scoring method in a voice evaluation scene, a storage medium and equipment. The method comprises the following steps: acquiring a user answer text and a reference answer text; performing clause processing on the reference answer text to obtain a reference answer clause set; according to the reference answer clause set, performing clause processing on the user answer text to obtain a user answer clause set; traversing the reference answer clause set, performing similarity retrieval in the user answer clause set, retaining sentence pairs of which the similarity is greater than a threshold T, and obtaining a reference answer and user answer matching sentence pair set; calculating the semantic similarity of each sentence pair in a sentence pair set matched with the reference answer and the user answer; and performing weighted calculation on the semantic similarity of each sentence pair to obtain a total score. By using the method of the invention, the scoring stability and interpretability of the semi-open question type in the voice evaluation scene can be improved.

Description

Technical Field

[0001] The present invention relates to the field of intelligent education technology, and in particular to a scoring method, storage medium, and device for semi-open questions in a speech assessment scenario. Background Art

[0002] Semi-open questions in speech assessments are exam scenarios where the responses to spoken language are diverse and the reference answers cannot cover all of them. In these scenarios, the diversity of user expressions makes it difficult for their answers to be completely consistent with the reference answers. Currently, automatic scoring for speech assessments typically involves first recognizing the user's speech and then matching the recognition results with keywords in the reference answers to assign a score. However, this method is not stable and interpretable when applied to semi-open questions. Summary of the Invention

[0003] In view of the shortcomings of the existing technology, the present invention aims to provide a semi-open question scoring method, storage medium and device in a speech evaluation scenario.

[0004] In order to achieve the above object, the present invention adopts the following technical solutions:

[0005] A scoring method for semi-open questions in a speech assessment scenario includes the following steps:

[0006] S1. Obtain the user's answer text through voice recognition and obtain the reference answer text;

[0007] S2. Segment the reference answer text to obtain a reference answer sentence set;

[0008] S3. Segment the user answer text according to the sentence segmentation result of the reference answer text in step S2 to obtain a set of user answer sentences;

[0009] S4, traversing the reference answer sentence set obtained in step S2, performing a similarity search on each reference answer sentence in the reference answer sentence set with the user answer sentence set obtained in step S3, and retaining sentence pairs with a similarity greater than a threshold T, thereby obtaining a set of matching reference answer and user answer sentence pairs;

[0010] S5. Calculate semantic similarity for each sentence pair in the set of user answer matching sentence pairs obtained in step S4;

[0011] S6. Perform weighted summation on the semantic similarity of each sentence pair in the set of sentence pairs matching the user's answer to obtain the total score of the user's answer.

[0012] Furthermore, in step S2 and step S3, the reference answer text and the user answer text are respectively segmented using the BERT sequence labeling model, and in step S3, the input of the BERT sequence labeling model for segmenting the user answer text is the concatenation of the sentence segmentation results of the reference answer text and the user answer text.

[0013] Furthermore, in step S4, the similarity search uses Sentence-BERT sentence vector search.

[0014] Furthermore, in step S5, the semantic similarity of the sentence pairs is calculated using the Cross-Encoder model of Sentence-BERT.

[0015] The present invention also provides a computer-readable storage medium, characterized in that a computer program is stored in the computer-readable storage medium, and the computer program implements the above method when executed by a processor.

[0016] The present invention also provides a computer device, comprising a processor and a memory, wherein the memory is used to store a computer program; and when the processor is used to execute the computer program, the above method is implemented.

[0017] The beneficial effects of the present invention are as follows: the method of the present invention can improve the shortcomings of the traditional scoring method based on keyword matching by introducing the calculation of semantic similarity. It is particularly suitable for automatic scoring of semi-open questions in speech evaluation where users' answers are diverse and the keyword expression range is limited and difficult to enumerate. It can effectively improve the scoring stability and interpretability in such scenarios. DETAILED DESCRIPTION

[0018] The present invention will be further described below. It should be noted that this embodiment is based on the technical solution and provides a detailed implementation method and specific operation process, but the protection scope of the present invention is not limited to this embodiment.

[0019] This embodiment provides a scoring method for semi-open questions in a speech assessment scenario, including the following steps:

[0020] S1. Obtain the user's answer text through voice recognition and obtain the reference answer text.

[0021] S2. Segment the reference answer text to obtain a reference answer sentence set. In this embodiment, the BERT sequence tagging model is used to segment the reference answer text.

[0022] S3. Based on the sentence segmentation results of the reference answer text in step S2, the user answer text is segmented to obtain a set of user answer sentences.

[0023] In this embodiment, corresponding to the sentence segmentation method of the reference answer text, the BERT sequence labeling model is also used to sentence the user answer text. The input of the BERT sequence labeling model for sentence segmentation of the user answer text is the concatenation of the sentence segmentation result of the reference answer text and the user answer text. The sentence segmentation result of the reference answer text and the user answer text are separated by [SEP].

[0024] It should be noted that the principle of the BERT sequence tagging model that transforms sentence segmentation into a sequence tagging problem is as follows:

[0025] Input: original text

[0026] Tags: B-SEG (beginning of sentence), I-SEG (middle of sentence), O (non-sentence boundary)

[0027] For example,

[0028] Original text: "Hello! What a nice day today. Do you want to go out?"

[0029] Tags: Hello! Today... Do you want to...?

[0030] Finally, B-SEG tags are used to segment sentences.

[0031] S4. Traverse the set of reference answer sentences obtained in step S2, perform a similarity search on each reference answer sentence in the set of user answer sentences obtained in step S3, and retain sentence pairs with a similarity greater than a threshold T, thereby obtaining a set of matching reference answer and user answer sentence pairs. In this embodiment, the similarity search uses Sentence-BERT sentence vector search.

[0032] S5. Calculate semantic similarity for each sentence pair in the set of user answer matching sentence pairs obtained in step S4. In this embodiment, the semantic similarity of the sentence pairs is calculated using the Cross-Encoder model of Sentence-BERT.

[0033] S6. Perform weighted summation of the semantic similarity of each sentence pair in the set of matching sentence pairs of the user's answer to obtain the total score of the user's answer. In this embodiment, the weight of each sentence pair is the same, that is, W = 1 / total number of sentence pairs.

[0034] Those skilled in the art can make various corresponding changes and modifications based on the above technical solutions and concepts, and all of these changes and modifications should be included in the scope of protection of the claims of the present invention.

Claims

1. A scoring method for semi-open questions in a speech assessment scenario, characterized by: The steps include: S1. Obtain the user's answer text through voice recognition and obtain the reference answer text; S2. Segment the reference answer text to obtain a reference answer sentence set; S3. Segment the user answer text according to the sentence segmentation result of the reference answer text in step S2 to obtain a set of user answer sentences; S4, traversing the reference answer sentence set obtained in step S2, performing a similarity search on each reference answer sentence in the reference answer sentence set with the user answer sentence set obtained in step S3, and retaining sentence pairs with a similarity greater than a threshold T, thereby obtaining a set of matching reference answer and user answer sentence pairs; S5. Calculate semantic similarity for each sentence pair in the set of user answer matching sentence pairs obtained in step S4; S6. Perform weighted summation on the semantic similarity of each sentence pair in the set of sentence pairs matching the user's answer to obtain the total score of the user's answer.

2. The method according to claim 1, characterized in that In steps S2 and S3, the reference answer text and the user answer text are respectively segmented using the BERT sequence labeling model, and in step S3, the input of the BERT sequence labeling model for segmenting the user answer text is the concatenation of the segmentation results of the reference answer text and the user answer text.

3. The method according to claim 1, characterized in that In step S4, the similarity search uses Sentence-BERT sentence vector search.

4. The method according to claim 1, wherein In step S5, the semantic similarity of the sentence pairs is calculated using the Cross-Encoder model of Sentence-BERT.

5. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the method according to any one of claims 1 to 4 is implemented.

6. A computer device, characterized in that: The method comprises a processor and a memory, wherein the memory is used to store a computer program; and when the processor is used to execute the computer program, the method according to any one of claims 1 to 4 is implemented.