Automatic reviewing method and device for subjective questions

Abstract

The invention relates to a subjective question automatic marking method and device, and the method comprises the steps: obtaining a reference answer and an examinee answer, and carrying out the preprocessing of the reference answer and the examinee answer through a natural language processing technology, and obtaining a preprocessing result; the preprocessing result is converted on the basis of a pre-trained SenteceTransform model, and a reference answer vector and an examinee answer vector are obtained; according to a synonym forest traversal preprocessing result, determining a keyword hit result; calculating the similarity between the reference answer vector and the examinee answer vector to obtain a similarity result; and presetting a scoring standard, and determining a scoring result in combination with the similarity result and the keyword hit result. Vector calculation is carried out on reference answers and examinee answers through a SentenceTransform model, semantic information of answer sentences is captured, a scoring result is determined according to a similarity result and a keyword hit result by referring to a scoring standard, the examinee answers are rapidly scored, the influence of human subjective expression inconsistency on an evaluation result is effectively avoided, and the evaluation efficiency is improved. And the scoring result is objective and accurate.

Description

Technical Field

[0001] The present disclosure relates to the technical field of natural language processing, and in particular to a method and device for automatically marking subjective questions. Background Art

[0002] In traditional education assessment, subjective questions are usually scored manually by teachers, which is not only time-consuming and inefficient, but also easily affected by human subjective factors, making it difficult to provide accurate and real-time scoring feedback. Moreover, the expressions of candidates for subjective questions are also subjective, and different candidates may express the same semantic keyword differently. Therefore, conventional keyword matching methods are limited by the diversity of natural language and often have low accuracy, which affects the scoring results.

[0003] Therefore, providing an automatic marking method for subjective questions to improve the efficiency and accuracy of scoring subjective questions is an urgent problem to be solved. Summary of the invention

[0004] In view of this, the present application proposes a method and device for automatically marking subjective questions to solve the above problems.

[0005] In one aspect, the present application proposes a method for automatically marking subjective questions, comprising the following steps:

[0006] Obtaining reference answers and examinee answers, and preprocessing the reference answers and examinee answers using natural language processing technology to obtain preprocessing results;

[0007] The preprocessing result is transformed based on the pre-trained SentenceTransformer model to obtain a reference answer vector and a candidate answer vector;

[0008] Determine the keyword hit result according to the preprocessing result of the synonym forest traversal;

[0009] Calculate the similarity between the reference answer vector and the examinee's answer vector to obtain a similarity result;

[0010] A scoring standard is preset, and a scoring result is determined in combination with the similarity result and the keyword hit result.

[0011] As an optional implementation scheme of the present application, optionally, obtaining reference answers and examinee answers, and preprocessing the reference answers and examinee answers using natural language processing technology to obtain preprocessing results, including:

[0012] Get reference answers and corresponding examinee answers;

[0013] Perform word segmentation processing on the reference answer and the examinee's answer respectively to obtain a word segmentation result;

[0014] Removing stop words from the word segmentation result;

[0015] The word segmentation result after the stop words are removed is subjected to part-of-speech restoration to obtain a preprocessing result.

[0016] As an optional implementation scheme of the present application, optionally, the preprocessing result is transformed based on a pre-trained SentenceTransformer model to obtain a reference answer vector and a candidate answer vector, including:

[0017] Obtaining the preprocessing result;

[0018] Training the SentenceTransformer model by learning semantic relations in a text corpus;

[0019] The trained SentenceTransformer model is used to encode the semantic information in the preprocessing result into a reference answer vector and a candidate answer vector respectively.

[0020] As an optional implementation scheme of the present application, optionally, the semantic information in the preprocessing result is encoded into a reference answer vector and a candidate answer vector respectively using the trained SentenceTransformer model, including:

[0021] Acquiring semantic information of the preprocessing result;

[0022] Encoding the semantic information using an encode method to obtain an encoding result;

[0023] The encoding results are extracted to obtain a reference answer vector and a candidate answer vector.

[0024] As an optional implementation scheme of the present application, optionally, the reference answer vector and the examinee answer vector are both 1024-dimensional vectors.

[0025] As an optional implementation scheme of the present application, optionally, when calculating the similarity between the reference answer vector and the examinee answer vector and obtaining the similarity result, it is implemented based on a cosine similarity algorithm.

[0026] As an optional implementation scheme of the present application, optionally, after presetting the scoring criteria and combining the similarity result and the keyword hit result to determine the scoring result, it includes:

[0027] Feedback information is generated using the scoring result.

[0028] In another aspect, the present application provides a device for implementing any of the above-mentioned methods for automatically marking subjective questions, comprising:

[0029] A data preprocessing module is configured to obtain reference answers and examinee answers, and preprocess the reference answers and examinee answers using natural language processing technology to obtain preprocessing results;

[0030] A sentence vector calculation module is configured to transform the preprocessing result based on a pre-trained SentenceTransformer model to obtain a reference answer vector and a candidate answer vector;

[0031] A keyword matching module is configured to determine a keyword hit result by traversing the preprocessing result according to the synonym forest;

[0032] A similarity calculation module is configured to calculate the similarity between the reference answer vector and the examinee answer vector by using a cosine similarity algorithm to obtain a similarity result;

[0033] The scoring module is configured to preset scoring criteria and determine the scoring result in combination with the similarity result.

[0034] In another aspect of the present application, an electronic device is provided, comprising: a processor;

[0035] a memory for storing processor-executable instructions;

[0036] Wherein, the processor is configured to implement any of the above-mentioned methods for automatically marking subjective questions when executing the executable instructions.

[0037] On another aspect of the present application, a non-volatile computer-readable storage medium is provided, on which computer program instructions are stored. When the computer program instructions are executed by a processor, any one of the above-mentioned methods for automatically marking subjective questions is implemented.

[0038] Technical effects of the present invention:

[0039] The automatic marking method for subjective questions of the present application includes obtaining reference answers and examinee answers, and preprocessing the reference answers and examinee answers using natural language processing technology to obtain preprocessing results; transforming the preprocessing results based on a pre-trained SentenceTransformer model to obtain reference answer vectors and examinee answer vectors; determining keyword hit results based on traversing the preprocessing results through a synonym forest; calculating the similarity between the reference answer vector and the examinee answer vector to obtain a similarity result; presetting a scoring standard, and determining the scoring result in combination with the similarity result and the keyword hit result. The method of the present application uses a SentenceTransformer model to perform vector calculations on the preprocessed reference answers and examinee answers, respectively, to capture the semantic information in the answer sentences, and after combining the obtained similarity results between the reference answer vector and the examinee answer vector, and the keyword hit results obtained by traversing the preprocessing results, the scoring result of the examinee answer is determined with reference to the preset scoring standard. It can not only quickly and automatically score candidates' answers, reducing teachers' workload and saving their time and energy, but also vectorize the answers based on the Sentence Transformer model, effectively avoiding the impact of inconsistent subjective expressions on the evaluation results, making the scoring results more objective and accurate.

[0040] Further features and aspects of the present disclosure will become apparent from the following detailed description of exemplary embodiments with reference to the attached drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] The accompanying drawings, which are incorporated in and constitute a part of the specification, illustrate exemplary embodiments, features, and aspects of the disclosure and, together with the description, serve to explain the principles of the disclosure.

[0042] Figure 1 A flow chart of the method for automatically marking subjective questions of the present invention is shown;

[0043] Figure 2 Shown is a schematic diagram of the implementation process of the method for automatic marking of subjective questions of the present invention. DETAILED DESCRIPTION

[0044] Various exemplary embodiments, features and aspects of the present disclosure will be described in detail below with reference to the accompanying drawings. The same reference numerals in the accompanying drawings represent elements with the same or similar functions. Although various aspects of the embodiments are shown in the accompanying drawings, the drawings are not necessarily drawn to scale unless otherwise specified.

[0045] The word “exemplary” is used exclusively herein to mean “serving as an example, example, or illustration.” Any embodiment described herein as “exemplary” is not necessarily to be construed as preferred or advantageous over other embodiments.

[0046] In addition, in order to better illustrate the present disclosure, numerous specific details are given in the following specific embodiments. It should be understood by those skilled in the art that the present disclosure can also be implemented without certain specific details. In some examples, methods, means, components and circuits well known to those skilled in the art are not described in detail in order to highlight the subject matter of the present disclosure.

[0047] Example 1

[0048] like Figure 1 and Figure 2 As shown, in one aspect, the present application proposes a method for automatically marking subjective questions, comprising the following steps:

[0049] S100, obtaining reference answers and examinee answers, and preprocessing the reference answers and examinee answers using natural language processing technology to obtain preprocessing results;

[0050] S200, transforming the preprocessing result based on the pre-trained SentenceTransformer model to obtain a reference answer vector and a candidate answer vector;

[0051] S300, traversing the preprocessing result according to the synonym forest to determine the keyword hit result;

[0052] S400, calculating the similarity between the reference answer vector and the examinee's answer vector to obtain a similarity result;

[0053] S500: Preset a scoring standard, and determine a scoring result in combination with the similarity result and the keyword hit result.

[0054] In this embodiment, the SentenceTransformer model is used to calculate the vectors of the preprocessed reference answers and the examinee's answers, respectively, to capture the semantic information in the answer sentences, and the similarity results between the reference answer vector and the examinee's answer vector are combined with the keyword hit results obtained by traversing the preprocessing results, and then the score results of the examinee's answers are determined with reference to the preset scoring criteria. Not only can the examinee's answers be scored quickly and automatically, reducing the workload of teachers and saving their time and energy, but the answers are also represented by vectorization based on the SentenceTransformer model, effectively avoiding the impact of inconsistent subjective expressions on the evaluation results, making the scoring results more objective and accurate.

[0055] Specifically, through step S100, reference answers and examinee answers are obtained, and the reference answers and examinee answers are preprocessed using natural language processing technology to obtain preprocessing results. By collecting a large number of subjective question answer data sets, including examinee answers and corresponding standard answers, it is ensured that the data sets are diverse and representative so that they can cover various types of questions and answers. Among them, the data preprocessing module uses natural language processing technology, namely NLP technology, to preprocess the reference answers and examinee answers respectively, wherein the preprocessing includes removing stop words, word segmentation and part of speech restoration, thereby reducing noise and extracting meaningful features in the answers, which is convenient for subsequent vectorization processing.

[0056] As an optional implementation scheme of the present application, optionally, obtaining reference answers and examinee answers, and preprocessing the reference answers and examinee answers using natural language processing technology to obtain preprocessing results, including:

[0057] Get reference answers and corresponding examinee answers;

[0058] Perform word segmentation processing on the reference answer and the examinee's answer respectively to obtain a word segmentation result;

[0059] Removing stop words from the word segmentation result;

[0060] The word segmentation result after the stop words are removed is subjected to part-of-speech restoration to obtain a preprocessing result.

[0061] That is, after obtaining the reference answer and the corresponding candidate answer, first perform Chinese word segmentation processing on them, and split the sentences in the answer into independent words. For example, the candidate's answer to a subjective question in the construction industry is: "The accident in this case should be characterized as a major safety accident. The three-level education is enterprise, project, and team education." In the process of word segmentation, the Chinese word segmentation tool jieba is used to segment the above sentence into: "In this case", "accident", "should", "qualitative", "for", "major safety accident", "three-level education", "for", "enterprise", "project", "team education". Further, the stop words are removed from the word segmentation results after word segmentation processing. Since stop words are words that appear frequently in the text but lack actual meaning, they need to be removed during the preprocessing of the reference answer and the candidate's answer. Preferably, a stop word list is used to remove the stop words in the word segmentation results. For example, for the above word segmentation results, stop words such as "this", "in", "should", and "for" are removed. Furthermore, the word segmentation results after removing stop words are restored to their parts of speech. Specifically, part of speech restoration in Chinese is the process of restoring a word to its basic form. For the word segmentation results after removing stop words, since there are no complex Chinese part of speech change rules, explicit part of speech restoration processing is not required, that is, the final result after preprocessing is: "Case accident qualitative major safety accident level 3 education enterprise project team education".

[0062] After obtaining the preprocessing result, the preprocessing result is transformed based on the pre-trained SentenceTransformer model through step S200 to obtain a reference answer vector and a candidate answer vector.

[0063] Among them, as an optional implementation scheme of the present application, optionally, the preprocessing result is transformed based on the pre-trained SentenceTransformer model to obtain a reference answer vector and a candidate answer vector, including:

[0064] Obtaining the preprocessing result;

[0065] Training the SentenceTransformer model by learning semantic relations in a text corpus;

[0066] The trained SentenceTransformer model is used to encode the semantic information in the preprocessing result into a reference answer vector and a candidate answer vector respectively.

[0067] Further, as an optional implementation scheme of the present application, optionally, the semantic information in the preprocessing result is encoded into a reference answer vector and a candidate answer vector respectively using the trained SentenceTransformer model, including:

[0068] Acquiring semantic information of the preprocessing result;

[0069] Encoding the semantic information using an encode method to obtain an encoding result;

[0070] The encoding results are extracted to obtain a reference answer vector and a candidate answer vector.

[0071] In this embodiment, the input sentence is converted into a vector representation using the SentenceTransformer model. It should be noted that SentenceTransformer, as a pre-trained semantic model, is trained by learning the semantic relationship in a large-scale corpus. During the training process, it learns to capture the semantic information of the sentence and encodes it into a high-dimensional vector. Specifically, for the pre-processing result, the trained Chinese "bert-base-chinese" model is first loaded through the sentence vector calculation module, and then the pre-processed sentence is input into the loaded SentenceTransformer model to obtain the encoding of the sentence, wherein after the input sentence is passed to the model's encode method, a sentence encoding encoded_sentence is returned, and then the sentence vector is extracted from the encoding result to ensure that the vector retains the semantic information of the answer. It should be noted that when using the pre-trained SentenceTransformer model, the vector in the first position is usually used as the sentence vector, thereby obtaining the vector sentence_vector representing the sentence. It should also be noted that as an optional implementation scheme of the present application, optionally, the reference answer vector and the candidate answer vector are both 1024-dimensional vectors.

[0072] Through step S300, the preprocessing results are traversed according to the synonym forest to determine the keyword hit results. It should be noted that before the candidate's answer is scored, the keyword hit results are determined according to the preprocessing results. Specifically, through the keyword matching module, on the basis of the Harbin Institute of Technology word forest, a subject-specific academic synonym forest is added, and the phrase results obtained after the preprocessing of the candidate's answer and the reference answer are traversed one by one to obtain the keyword hit results of the candidate's answer and the reference answer. The method of this embodiment, for the terms collected in the synonym forest, the keywords will also be hit even if they are not completely matched. For example, the candidate's answer is "third-level education company project department team", and the reference answer is "third-level education enterprise project team", then the four words of the candidate's answer will all be hit, that is, it can effectively avoid the problem of inaccurate scoring results caused by human subjectivity.

[0073] Through step S400, the similarity between the reference answer vector and the examinee's answer vector is calculated to obtain a similarity result;

[0074] As an optional implementation scheme of the present application, optionally, when calculating the similarity between the reference answer vector and the examinee answer vector and obtaining the similarity result, it is implemented based on a cosine similarity algorithm.

[0075] In this embodiment, for the reference answer vector and the candidate answer vector output by the sentence vector module, the similarity calculation module uses the cosine similarity algorithm to calculate the similarity between the candidate answer vector and the reference answer vector according to the cosine similarity calculation formula. The higher the similarity, the closer the candidate's answer is to the standard reference answer, and the higher the score.

[0076] After determining the similarity results and keyword hit results, step S500 is performed, the scoring criteria are preset, and the scoring results are determined in combination with the similarity results and the keyword hit results. Among them, the similarity results between the input candidate answers and the reference answers, the keyword hit results, that is, the keyword matching percentage, and the relatively objective score percentage (the score value of 1 point out of full marks) are combined through the scoring module. The scoring module presets the scoring rules for each question in advance by the teacher. After loading the scoring model and scoring rules, the similarity results and keyword hit results are used as input, converted into specific scores, and the scoring results are generated. For the scoring criteria, teachers can customize them as needed to adapt to different types of questions and answers.

[0077] As an optional implementation scheme of the present application, optionally, after presetting the scoring criteria and combining the similarity result and the keyword hit result to determine the scoring result, it includes:

[0078] Feedback information is generated using the scoring result.

[0079] It should be noted that after the scoring result is determined by the scoring module, feedback information is generated by using the scoring result through step S600, that is, detailed feedback information is generated according to the scoring result through the feedback module, including scoring explanation, error analysis and guidance for improvement. For example, the hit keywords are marked in green, and the short sentences with high similarity to the answer are marked in blue. Candidates can get instant feedback through the system to understand their scores and improvement directions. At the same time, the overall scoring accuracy and stability of the system can be improved by introducing more training data or adjusting the similarity calculation algorithm.

[0080] Specific examples are provided below for further explanation.

[0081] Answer: When choosing a manufacturer, the following factors should be considered: (1) The supplier’s geographical location. (2) Technology and production capacity. (3) The sufficiency of production tasks. (4) The supplier’s reputation.

[0082] Candidate's answer: The main considerations for choosing a manufacturer are: location, technical productivity, and credit

[0083] Scoring criteria: Geographic location #%#2 points; Technology and production capacity #%#2 points; Fullness of production tasks #%#1 point; Reputation #%#1 point

[0084] 1. Data preprocessing

[0085] (1) Word segmentation: Use the Chinese word segmentation tool (jieba) to segment the sentence into independent words. The reference answer after word segmentation is: "in", "select", "manufacturer", "when", "mainly", "consider", "the following", "several", "aspects", "of", "factors", ":", "(1)", "supplier", "of", "geographic location", ".", "(2)", "technology", "、", "production capacity", ".", "(3)", "production task", "of", "fullness", ".", "(4)", "supplier", "of", "reputation", ".".

[0086] The results of the candidates' answers are: "right", "this sentence", "conduct", "data preprocessing", ",", "select", "manufacturer", "mainly", "consider", ":", "position", ",","technical productivity", ",","credit".

[0087] (2) Remove stop words: Use a stop word list to remove words that appear frequently in the text but lack actual meaning. The processing results of the reference answers are: "select", "manufacturer", "supplier", "geographic location", "technology", "production capacity", "production task", "fullness", "supplier", "reputation"

[0088] The processing results of the candidates' answers are: "select", "manufacturer", "location", "technology", "productivity", and "credit".

[0089] 2. Keyword hits

[0090] Based on the synonym forest and preprocessing results, the keyword hit results are: "selection", "manufacturer", "location", "technology", "productivity", "credit", and the hit rate is 0.6.

[0091] 3. Use the SentenceTransformer model to calculate the sentence vectors of the reference answer and the student's answer, encoded_sentence = bert_model.encode([sentence]), and get two 1024-dimensional vectors.

[0092] 4. The similarity sim calculated by the cosine similarity formula is approximately 0.819532.

[0093] 5. After loading the model, call the score generation function result = mark_model.mark(sim, idf, std) to generate a score of 4.9.

[0094] 6. Result feedback: The main considerations for selecting manufacturers are #%#0.9: location #%#1.0, technical productivity #%#1.0, credit #%#1.0#%#4.9#%#The fullness of the production task is omitted.

[0095] In summary, the automatic marking method for subjective questions of the present application can quickly and automatically score the answers of candidates, reduce the workload of teachers, and save time and energy. At the same time, due to the use of semantic models for vectorized representation, the scoring results are more accurate and objective. Not only that, candidates can obtain scoring results and detailed feedback information in real time to understand their scores and directions for improvement. This helps to promote students' learning progress and improve their self-learning ability. Since the system provides flexible scoring criteria and feedback generation mechanisms, teachers can customize scoring rules as needed to adapt to different types of questions and answers. In addition, by continuously introducing more training data and optimizing model algorithms, the scoring accuracy and stability of the system can be continuously improved, providing a more intelligent solution for educational evaluation. Therefore, the automatic marking method for subjective questions based on SentenceTransformer provided in the present application can improve evaluation efficiency and accuracy while realizing automatic scoring and timely feedback of subjective questions in the field of education.

[0096] It should be noted that, although the above is introduced as an example, those skilled in the art will understand that the present disclosure should not be limited thereto. In fact, users can flexibly set according to actual application scenarios, as long as the technical functions of the present application can be realized according to the above technical methods.

[0097] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment method can be implemented by instructing the relevant hardware through a computer program, and the program can be stored in a computer-readable storage medium. When the program is executed, it can include the processes of the embodiments of the above-mentioned control methods. Among them, the storage medium can be a disk, an optical disk, a read-only memory (ROM), a random access memory (RAM), a flash memory (Flash Memory), a hard disk (Hard Disk Drive, abbreviated as: HDD) or a solid-state drive (SSD), etc.; the storage medium can also include a combination of the above-mentioned types of memory.

[0098] Example 2

[0099] On the other hand, the present application provides a device for implementing any of the above-mentioned methods for automatically marking subjective questions. Since the working principle of the device in the embodiment of the present disclosure is the same or similar to the principle of the method for automatically marking subjective questions in the embodiment of the present disclosure, the repeated parts will not be repeated. The device in the embodiment of the present application includes:

[0100] A data preprocessing module is configured to obtain reference answers and examinee answers, and preprocess the reference answers and examinee answers using natural language processing technology to obtain preprocessing results;

[0101] A sentence vector calculation module is configured to transform the preprocessing result based on a pre-trained SentenceTransformer model to obtain a reference answer vector and a candidate answer vector;

[0102] A keyword matching module is configured to traverse the preprocessing result according to the synonym forest to determine the keyword hit result;

[0103] A similarity calculation module is configured to calculate the similarity between the reference answer vector and the examinee answer vector by using a cosine similarity algorithm to obtain a similarity result;

[0104] The scoring module is configured to preset scoring criteria and determine the scoring result in combination with the similarity result.

[0105] Example 3

[0106] Furthermore, another aspect of the present application provides an electronic device, including: a processor;

[0107] a memory for storing processor-executable instructions;

[0108] Wherein, the processor is configured to implement any of the above-mentioned methods for automatically marking subjective questions when executing the executable instructions.

[0109] The control system of the embodiment of the present disclosure includes a processor and a memory for storing processor executable instructions. The processor is configured to implement any of the above-mentioned methods for automatically marking subjective questions when executing the executable instructions.

[0110] Here, it should be noted that the number of processors can be one or more. At the same time, in the control system of the embodiment of the present disclosure, an input device and an output device may also be included. Among them, the processor, memory, input device and output device may be connected through a bus or in other ways, which are not specifically limited here.

[0111] The memory, as a computer-readable storage medium, can be used to store software programs, computer executable programs and various modules, such as the program or module corresponding to the method for automatically marking subjective questions in the embodiment of the present disclosure. The processor executes various functional applications and data processing of the control system by running the software programs or modules stored in the memory.

[0112] The input device can be used to receive input numbers or signals. The signal can be a key signal related to user settings and function control of the device / terminal / server. The output device can include a display device such as a display screen.

[0113] Example 4

[0114] On another aspect of the present application, a non-volatile computer-readable storage medium is provided, on which computer program instructions are stored. When the computer program instructions are executed by a processor, any one of the above-mentioned methods for automatically marking subjective questions is implemented.

[0115] The embodiments of the present disclosure have been described above, and the above description is exemplary, not exhaustive, and is not limited to the disclosed embodiments. Many modifications and variations will be apparent to those of ordinary skill in the art without departing from the scope and spirit of the described embodiments. The selection of terms used herein is intended to best explain the principles of the embodiments, practical applications, or technical improvements to the technology in the market, or to enable other persons of ordinary skill in the art to understand the embodiments disclosed herein.

Claims

1. A method for automatically marking subjective questions, characterized in that: The steps include: Obtaining reference answers and examinee answers, and preprocessing the reference answers and examinee answers using natural language processing technology to obtain preprocessing results; The preprocessing result is transformed based on the pre-trained SentenceTransformer model to obtain a reference answer vector and a candidate answer vector; According to the pre-processing result of the synonym forest traversal, a keyword hit result is determined; Calculate the similarity between the reference answer vector and the examinee's answer vector to obtain a similarity result; A scoring standard is preset, and a scoring result is determined in combination with the similarity result and the keyword hit result.

2. The method for automatically marking subjective questions according to claim 1, characterized in that: Obtain reference answers and examinee answers, and preprocess the reference answers and examinee answers using natural language processing technology to obtain preprocessing results, including: Get reference answers and corresponding examinee answers; Perform word segmentation processing on the reference answer and the examinee's answer respectively to obtain a word segmentation result; Removing stop words from the word segmentation result; The word segmentation result after the stop words are removed is subjected to part-of-speech restoration to obtain a preprocessing result.

3. The method for automatically marking subjective questions according to claim 1, characterized in that: The preprocessing results are transformed based on the pre-trained SentenceTransformer model to obtain a reference answer vector and a candidate answer vector, including: Obtaining the preprocessing result; Training the SentenceTransformer model by learning semantic relations in a text corpus; The trained SentenceTransformer model is used to encode the semantic information in the preprocessing result into a reference answer vector and a candidate answer vector respectively.

4. The method for automatically marking subjective questions according to claim 3 is characterized in that: The semantic information in the preprocessing result is encoded into a reference answer vector and a candidate answer vector respectively using the trained SentenceTransformer model, including: Acquiring semantic information of the preprocessing result; Encoding the semantic information using an encode method to obtain an encoding result; The encoding results are extracted to obtain a reference answer vector and a candidate answer vector.

5. The method for automatically marking subjective questions according to claim 4 is characterized in that: The reference answer vector and the examinee answer vector are both 1024-dimensional vectors.

6. The method for automatically marking subjective questions according to claim 1, characterized in that: When calculating the similarity between the reference answer vector and the examinee answer vector and obtaining the similarity result, it is implemented based on the cosine similarity algorithm.

7. The method for automatically marking subjective questions according to any one of claims 1 to 6, characterized in that: After the scoring criteria are preset and the scoring result is determined by combining the similarity result and the keyword hit result, it includes: Feedback information is generated using the scoring result.

8. A device for implementing the method for automatically marking subjective questions as claimed in any one of claims 1 to 7, characterized in that: include: A data preprocessing module is configured to obtain reference answers and examinee answers, and preprocess the reference answers and examinee answers using natural language processing technology to obtain preprocessing results; A sentence vector calculation module is configured to transform the preprocessing result based on a pre-trained SentenceTransformer model to obtain a reference answer vector and a candidate answer vector; A keyword matching module is configured to determine a keyword hit result by traversing the preprocessing result according to the synonym forest; A similarity calculation module is configured to calculate the similarity between the reference answer vector and the examinee answer vector by using a cosine similarity algorithm to obtain a similarity result; The scoring module is configured to preset scoring criteria and determine the scoring result in combination with the similarity result.

9. An electronic device, characterized in that: include: processor; a memory for storing processor-executable instructions; Wherein, the processor is configured to implement the method for automatic marking of subjective questions according to any one of claims 1 to 7 when executing the executable instructions.

10. A non-volatile computer-readable storage medium having computer program instructions stored thereon, characterized in that: When the computer program instructions are executed by a processor, the method for automatically marking subjective questions as described in any one of claims 1 to 7 is implemented.

Images