A large model-based safety education courseware course generation method

Abstract

The application discloses a kind of based on big model's safety education courseware course generation method, it is related to big model technical field, including: obtaining the education content of the safety education courseware currently explained, call the historical generation record of safety education courseware, the compliance degree of each courseware template currently obtained, and generate courseware template recommendation list;The current education content includes several explanation blocks, obtain historical safety accident case, obtain the correlation degree of each safety accident case in each explanation block, generate the safety accident case list corresponding to each explanation block;Select target courseware template, generate safety education courseware course, in the process of explanation, relevant case is recommended to audience.The application avoids the problem that artificial selection is inefficient and case is split from teaching content by intelligently recommending courseware template and safety accident case, helps to improve safety education warning effect, achieves the effect of effectively effective safety warning and risk prevention.

Description

Technical Field

[0001] This invention relates to the field of large-scale model technology, specifically a method for generating safety education courseware based on large-scale models. Background Technology

[0002] Safety education courseware is an important safety education guide to ensure that people can respond to emergencies in a standardized manner during the workplace. It can effectively guide workers to deal with various sudden safety situations and avoid causing greater impact. However, in the current production process of safety education courseware, the instructor usually selects courseware templates and safety cases based on personal experience, without fully considering the personalized needs of the audience, such as their work scenarios. This results in courseware templates lacking scenario relevance. Furthermore, since the degree of matching between safety cases and the content of each section of the courseware varies, if cases are selected arbitrarily based solely on human experience, it is easy to create a disconnect between the cases and the teaching content. This leads to low safety education warning effects and makes it difficult to play a truly effective role in safety warning and risk prevention. Summary of the Invention

[0003] The purpose of this invention is to provide a method for generating safety education courseware based on a large model, so as to solve the problems raised in the prior art.

[0004] To solve the above-mentioned technical problems, the present invention provides the following technical solution: A method for generating safety education courseware based on a large model includes the following steps: Obtain the educational content corresponding to the current safety education courseware, as well as the target audience for this educational course; the safety education courseware here is for the onboarding training of relevant staff. For places prone to safety accidents such as factories and construction sites, safety education and training is of paramount importance. Only by discovering safety hazards and taking corresponding measures in a timely manner can we effectively ensure personal and property safety. Retrieve the historical generation records of safety education courseware, extract the corresponding educational content and courseware templates, analyze the current educational content and the educational content of the generation records, obtain the degree of conformity of each courseware template, and generate a recommended list of courseware templates. The current educational content includes several explanatory sections. Historical safety accident cases are obtained. Based on the current educational content and the target audience, the relevance of each safety accident case to each explanatory section is determined, and a list of safety accident cases corresponding to each explanatory section is generated. The presenter selects a target courseware template from the recommended list. Each corresponding section of the target courseware template contains a case viewing entry point, generating a safety education courseware lesson. During the presentation, the presenter recommends relevant cases to the audience.

[0005] Preferably, a recommended list of courseware templates is generated, including: Extract the educational content and courseware template M corresponding to a certain generated record, and establish a knowledge graph X1 for that educational content; obtain the educational content corresponding to the currently explained safety education courseware, and establish a knowledge graph X2 for that educational content; Extract the triples between knowledge graph X1 and knowledge graph X2, calculate the correlation between knowledge graph X1 and knowledge graph X2, and if the correlation is greater than the preset correlation threshold, then use the courseware template M as the target template. Several target templates are obtained, the corresponding tags of each target template are extracted, the suitability of each template is obtained from the current template selection platform, and a courseware template recommendation list is generated in descending order of suitability.

[0006] Preferably, calculating the correlation between knowledge graph X1 and knowledge graph X2 includes: A triple is a combination of keyword-relationship-keyword; the similarity between the first keyword, relation, and second keyword in any two triples is obtained, and the average is calculated to obtain the similarity between the two triples; By obtaining the number of triples N in knowledge graph X1 and the number of triples Q in knowledge graph X2, the correlation between knowledge graph X1 and knowledge graph X2 can be obtained. Where max() is used to find the maximum value, F n1 F n2 F nQ These are the similarities between the nth triplet in knowledge graph X1 and the 1st, 2nd, ..., Qth triplets in knowledge graph X2, respectively.

[0007] The greater the correlation, the more similar the two knowledge graphs are. In other words, the more similar the educational content corresponding to the generated record is to the educational content corresponding to the currently explained safety education courseware, the better the matching degree of the courseware template M corresponding to the generated record as the template of the currently explained safety education courseware. Therefore, in this scheme, if the correlation is greater than the preset correlation threshold, the courseware template M will be used as the more matching target template.

[0008] Preferably, the template conformity calculation steps include: extracting all tags corresponding to each target template, counting the occurrence frequency of each tag, normalizing the occurrence frequency, and using the normalization result as the distribution rate of each tag; obtaining all templates in the current template selection platform, summing the distribution rates of each tag corresponding to a certain template, and using the sum as the conformity of the template.

[0009] The greater the degree of conformity, the better the template is as a current safety education courseware template. Therefore, this solution obtains the degree of conformity of each courseware template in the current template selection platform, and generates a courseware template recommendation list in descending order of conformity, and intelligently recommends templates to speakers.

[0010] Preferably, a list of safety incident cases corresponding to each explanation section is generated, including: Obtain historical safety incident cases and extract case descriptions and affected areas corresponding to the safety incident cases; Obtain the current educational content, extract the section description corresponding to each lecture section, and determine the first correlation of each safety accident case in each lecture section based on the case description corresponding to the safety accident case. Obtain the target audience's work area, and determine the second relevance of each safety incident case based on the area of ​​influence corresponding to the safety incident case. Preset the weights of the first and second relevance, and based on the first and second relevance, obtain the degree of relevance of each safety incident case in each explanation section. Generate a list of safety incident cases corresponding to each explanation section in descending order of relevance.

[0011] Preferably, determining the primary correlation for each safety incident case includes: Extract all keywords from the description of a specific explanatory section to create a first keyword set; obtain all keywords from the description of a specific safety accident case to create a second keyword set; extract all keywords from the first and second keyword sets and remove duplicates to construct a third keyword set. First and second word vectors with the same dimensions as the third keyword set are established, with the initial value of each dimension being 0. Each keyword in the third keyword set is matched and statistically analyzed in the first and second keyword sets to obtain the final first and second word vectors. The cosine similarity between the two word vectors is calculated as the first relevance of the safety accident cases in this explanation section.

[0012] The greater the cosine similarity, the closer the two word vectors are in direction. In other words, the greater the cosine similarity, the greater the primary relevance of the safety accident case in the explanation section, which means that the case matches the section better. As a recommended case in the section, this case is more effective in explaining safety accidents to the audience and can better serve the purpose of safety warning. The specific calculation of cosine similarity is based on existing technology and will not be elaborated here.

[0013] Preferably, obtaining the final first and second word vectors includes: if the i-th keyword in the third keyword set exists in the first keyword set, then the value of the i-th dimension in the first word vector is modified to 1; if the j-th keyword in the third keyword set exists in the second keyword set, then the value of the j-th dimension in the second word vector is modified to 1.

[0014] Preferably, determining the second correlation for each safety incident case includes: Obtain the work areas of different area types within the workplace. Based on the number of people working in different work areas among the target audience and the area of ​​the work area of ​​the corresponding area type, obtain the area of ​​the work area corresponding to each area type. Then, normalize the area of ​​the work area and use the normalization result as the contribution ratio of each area type. The contribution ratio represents the importance and exposure of each region type in the target audience's work scenario. The larger the contribution ratio, the higher the proportion of that region type in the target audience's work scenario, the greater the degree of security risk exposure, and the more important it is to conduct security education.

[0015] Obtain the affected area involved when a case occurs, and extract the affected area of ​​different work areas within the affected area to obtain the second correlation of the safety accident case: Where H is the number of region types, W h S represents the contribution ratio of the h-th region type. h Let h be the area affected by the h-th region type.

[0016] The instructor selects a template from the recommended list of courseware templates based on actual needs. Each section of the generated safety education courseware course has a case viewing entry point. The case recommendation list for each section is arranged according to the list of safety accident cases. During the presentation, the instructor can select relevant cases according to actual needs and recommend them to the audience for learning. Combining high-quality cases with safety education and training can help improve the warning effect of safety education and achieve a practical and effective safety warning and risk prevention effect.

[0017] Compared with existing technologies, the beneficial effects of this invention are as follows: This invention provides a method for generating safety education courseware based on a large model, comprising: acquiring the educational content of the currently presented safety education courseware; retrieving the historical generation records of the safety education courseware; obtaining the conformity degree of each current courseware template; and generating a courseware template recommendation list; the current educational content includes several explanation modules; acquiring historical safety accident cases; obtaining the correlation degree of each safety accident case in each explanation module; and generating a safety accident case list corresponding to each explanation module; selecting a target courseware template; generating a safety education courseware; and recommending relevant cases to the audience during the presentation. This invention avoids the inefficiency of manual selection and the disconnect between cases and teaching content by intelligently recommending courseware templates and safety accident cases, thus helping to improve the safety education warning effect and achieving a truly effective safety warning and risk prevention function. Attached Figure Description

[0018] To more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 This is a flowchart illustrating a method for generating safety education courseware based on a large model, according to the present invention. Detailed Implementation

[0020] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numerals in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this disclosure. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this disclosure as detailed in the appended claims.

[0021] Example: Figure 1 As shown, this invention provides a technical solution for generating safety education courseware based on a large model, comprising the following steps: Obtain the educational content corresponding to the current safety education courseware, as well as the target audience for this educational course; the safety education courseware here is designed for the onboarding training of relevant personnel. For places prone to safety accidents, such as factories and construction sites, safety education and training is of paramount importance. Only by identifying safety hazards and taking corresponding measures in a timely manner can we effectively ensure personal and property safety.

[0022] Retrieve the historical generation records of safety education courseware. These records refer to the history of instructors selecting templates from the corresponding courseware template platform based on the training content and generating safety education courseware according to those templates. Extract the corresponding educational content and courseware templates from each generation record, analyze the current educational content and the content of the generated records to determine the suitability of each courseware template, and generate a recommended courseware template list. Specifically: Extract the educational content and courseware template M corresponding to a certain generated record, extract keywords and relationships from the educational content, and establish a knowledge graph X1 for the educational content; obtain the educational content corresponding to the safety education courseware currently being explained, extract keywords and relationships from the educational content, and establish a knowledge graph X2 for the educational content.

[0023] Extract triples between knowledge graph X1 and knowledge graph X2. Each triple is a combination of keyword-relationship-keyword. Calculate the correlation degree between knowledge graph X1 and knowledge graph X2. The correlation degree calculation includes the following steps: We obtain the similarity between the first keyword, the association, and the second keyword in any two triples, and calculate the average to obtain the similarity between the two triples. The similarity between words is obtained using existing large-scale model techniques, which will not be elaborated here. The formula for calculating the similarity between any two triples is as follows:

[0024] Where C1 is the similarity between the first keywords in the two triples, L is the similarity between the corresponding associations in the two triples, and C2 is the similarity between the second keywords in the two triples. By obtaining the number of triples N in knowledge graph X1 and the number of triples Q in knowledge graph X2, the correlation between knowledge graph X1 and knowledge graph X2 can be obtained.

[0025] Where max() is used to find the maximum value, F n1 F n2 F nQ The similarity is defined as the similarity between the nth triplet in knowledge graph X1 and the 1st, 2nd, ..., Qth triplets in knowledge graph X2. A higher similarity indicates a greater similarity between the two knowledge graphs, meaning a greater similarity between the educational content corresponding to the generated record and the educational content corresponding to the currently presented safety education courseware. Therefore, the better the matching degree of the courseware template M corresponding to the generated record as the template for the currently presented safety education courseware, the better. Thus, in this scheme, if the similarity is greater than the preset similarity threshold, the courseware template M is used as the more closely matched target template. Several target templates are obtained, and the corresponding tags for each target template are extracted. The suitability of each template is obtained from the current template selection platform. Based on the suitability in descending order, a recommended list of courseware templates is generated, including: Extract all tags corresponding to each target template, count the occurrences of each tag, and normalize the occurrences. Use the normalization result as the distribution rate of each tag. For example, if the tags are A, B, C, and D, count the occurrences of A, B, C, and D, and use the sigmoid normalization formula to normalize each occurrence. Use the normalization result as the distribution rate of A, B, C, and D, respectively.

[0026] To determine the suitability of a template, obtain all templates from the platform currently selected. Sum the distribution rates of each tag corresponding to a given template to determine the template's suitability. For example, if a template includes tags A, C, and D, then sum the distribution rates of tags A, C, and D to determine the template's suitability. The formula is as follows:

[0027] Among them, G i Let F represent the conformity of the i-th template, P be the number of tags in that template, and F be the degree of conformity. p Let P be the distribution rate of the p-th label. In this embodiment, P=3.

[0028] The greater the degree of conformity, the better the template is as a current safety education courseware template. Therefore, this solution obtains the degree of conformity of each courseware template in the current template selection platform, and generates a courseware template recommendation list in descending order of conformity, and intelligently recommends templates to speakers.

[0029] The current educational content includes several lecture modules, such as factory electrical safety, construction site traffic safety, and machinery and equipment safety. Historical safety accident cases are obtained, and based on the current educational content and the target audience, the relevance of each safety accident case to each lecture module is determined. This generates a list of safety accident cases corresponding to each lecture module, specifically: Obtain historical safety incident cases, extract the case descriptions and affected areas corresponding to the safety incident cases. The case descriptions are detailed descriptions of the causes, processes, consequences, and losses of the incidents based on the cases in question. The affected areas are the areas involved in the incidents. Obtain the current educational content, extract the section description corresponding to each lecture segment, and determine the first relevance of each safety accident case in each lecture segment based on the case descriptions corresponding to the safety accident cases. Specifically: Extract all keywords from the description of a specific explanation section to establish a first keyword set; obtain all keywords from the description of a specific safety accident case to establish a second keyword set; extract all keywords from the first and second keyword sets and remove duplicates to construct a third keyword set; assuming the first keyword set is {a, b, c, d} and the second keyword set is {a, d, e}, then based on all keywords from the first and second keyword sets and after removing duplicates, construct the third keyword set as {a, b, c, d, e}.

[0030] First and second word vectors with the same dimensions as the third keyword set are established respectively, wherein the initial value of each dimension is 0; in the embodiment, the dimension of the third keyword set is 5, so the first word vector is [0,0,0,0,0] and the second word vector is [0,0,0,0,0].

[0031] Each keyword in the third keyword set is matched and statistically analyzed in the first and second keyword sets respectively to obtain the final first and second word vectors. Specifically, if the i-th keyword in the third keyword set exists in the first keyword set, the value of the i-th dimension in the first word vector is modified to 1; if the j-th keyword in the third keyword set exists in the second keyword set, the value of the j-th dimension in the second word vector is modified to 1. For example, since the first keyword 'a' in the third keyword set exists in the first keyword set, the value of the first dimension in the first word vector is modified to 1. If the fifth keyword 'e' in the third keyword set does not exist in the first keyword set, the value of the fifth dimension in the first word vector remains unchanged, and so on. The final first word vector is [1,1,1,1,0], and the final second word vector is [1,0,0,1,1].

[0032] The cosine similarity between two word vectors is calculated. The larger the cosine similarity, the closer the two word vectors are in direction. In other words, the larger the cosine similarity, the greater the primary relevance of the safety accident case in the explanation section. This means that the case matches the section better, and the better it serves as a recommended case for the section, in explaining safety accidents to the audience and in achieving the purpose of safety warning. The specific calculation of cosine similarity is based on existing technology and will not be elaborated here.

[0033] Obtain the target audience's work area, and determine the second degree of relevance for each safety incident case based on the area of ​​influence corresponding to the case, as follows: The process involves identifying different work areas within the workplace, dividing it into different zones. Based on the number of people working in each zone among the target audience and the area of ​​the corresponding zone type, the work area for each zone type is calculated. This work area is then normalized, and the normalized result is used as the contribution ratio for each zone type. An example is shown below: For example, in this embodiment, the total area of ​​the workplace is 20,000 m². 2 The workplace includes various areas such as workshop areas, material storage areas, power distribution areas, and hazardous chemical areas. The total area of ​​the workshop area is 10,000 m². 2 The total area of ​​the material storage area is 4000m². 2 The total area of ​​the power distribution zone is 2000m². 2 The total area of ​​the hazardous chemicals zone is 4000m². 2 First, set the work area area for each area type to 0. Based on the number T1 of people in the target audience whose work area is the workshop area, the work area area corresponding to the workshop area is: T1 * 10000m² 2 By analogy, the working area corresponding to each area type can be obtained; The area of ​​the work area is normalized using the sigmoid formula. The normalization result is used as the contribution ratio of each area type. The contribution ratio represents the importance and exposure of each area type in the target audience's work scenario. The larger the contribution ratio, the higher the proportion of that area type in the target audience's work scenario, the greater the degree of safety risk exposure, and the more important it is to conduct safety education.

[0034] Obtain the affected area involved when a case occurs, and extract the affected area of ​​different work areas within the affected area to obtain the second correlation of the safety accident case:

[0035] Where H represents the number of region types, W h S represents the contribution ratio of the h-th region type. h Let h be the area affected by the h-th region type.

[0036] The second degree of relevance is obtained based on the contribution ratio and the area of ​​influence. The larger the second degree of relevance, the greater the significance of the case for the audience and the more likely it is to be recommended.

[0037] Preset weights for the first and second relevance degrees. Based on the first and second relevance degrees, the relevance degree of each safety incident case in each explanation section is obtained. In this embodiment, the weights V1 and V2 of the first and second relevance degrees are first set, with V1 being 0.6 and V2 being 0.4. According to the first relevance degree Cor1 of a certain safety incident case in a certain explanation section and the second relevance degree Rel of the safety incident case, the relevance degree of the safety incident case in the explanation section is obtained as: V1*Cor1+V2*Rel.

[0038] This allows us to obtain the correlation between multiple safety incident cases within the explanation section, and generate a list of safety incident cases for each explanation section in descending order of correlation. Each section has a corresponding list of safety incident cases. In this embodiment, each list of safety incident cases contains 5 cases, which are the top 5 cases with the highest correlation.

[0039] The presenter selects a target courseware template from the recommended list. Each corresponding section of the target courseware template contains a case viewing entry point, generating a safety education courseware lesson. During the presentation, the presenter recommends relevant cases to the audience based on actual needs.

[0040] The instructor selects a template from the recommended list of courseware templates based on actual needs. Each section of the generated safety education courseware course has a case viewing entry point. The case recommendation list for each section is arranged according to the list of safety accident cases. During the presentation, the instructor can select relevant cases according to actual needs and recommend them to the audience for learning. Combining high-quality cases with safety education and training can help improve the warning effect of safety education and achieve a practical and effective safety warning and risk prevention effect.

[0041] It should be noted that the order of the above embodiments of the present invention is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. Furthermore, specific embodiments have been described above. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps described in the claims can be performed in a different order than that shown in the embodiments and still achieve the desired result. Additionally, the processes depicted in the drawings do not necessarily require a specific or sequential order to achieve the desired result. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0042] The various embodiments in this specification are described in a progressive manner. The same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on describing the differences from other embodiments.

[0043] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for generating safety education courseware based on a large model, characterized in that, Includes the following steps: Obtain the educational content corresponding to the current safety education courseware, and the target audience for this educational course; Retrieve the historical generation records of safety education courseware, extract the corresponding educational content and courseware templates, analyze the current educational content and the educational content of the generation records, obtain the degree of conformity of each courseware template, and generate a recommended list of courseware templates. The current educational content includes several explanatory sections. Historical safety accident cases are obtained. Based on the current educational content and the target audience, the relevance of each safety accident case to each explanatory section is determined, and a list of safety accident cases corresponding to each explanatory section is generated. The presenter selects a target courseware template from the recommended list of courseware templates. Each corresponding section of the target courseware template contains a case viewing entry point, generating a safety education courseware course. During the presentation, the presenter recommends relevant cases to the audience. This includes generating a list of safety incident cases corresponding to each explanation section, including: Obtain historical safety incident cases and extract case descriptions and affected areas corresponding to the safety incident cases; Obtain the current educational content, extract the section description corresponding to each lecture section, and determine the first correlation of each safety accident case in each lecture section based on the case description corresponding to the safety accident case. Obtain the target audience's work area, and determine the second degree of relevance for each safety incident case based on the area of ​​impact corresponding to the safety incident case; Preset the weights of the first and second relevance degrees. Based on the first and second relevance degrees, obtain the degree of relevance of each safety accident case in each explanation section, and generate a list of safety accident cases corresponding to each explanation section in descending order of relevance degree. The determination of the second correlation degree for each safety incident case includes: Obtain the work areas of different area types within the workplace. Based on the number of people working in different work areas among the target audience and the area of ​​the work area of ​​the corresponding area type, obtain the area of ​​the work area corresponding to each area type. Then, normalize the area of ​​the work area and use the normalization result as the contribution ratio of each area type. Obtain the affected area involved when a case occurs, and extract the affected area of ​​different work areas within the affected area to obtain the second correlation of the safety accident case: Where H is the number of region types, W h S represents the contribution ratio of the h-th region type. h Let h be the area affected by the h-th region type.

2. The method for generating safety education courseware based on a large model according to claim 1, characterized in that, Generate a recommended list of courseware templates, including: Extract the educational content and courseware template M corresponding to a certain generated record, and establish a knowledge graph X1 for that educational content; obtain the educational content corresponding to the currently explained safety education courseware, and establish a knowledge graph X2 for that educational content; Extract the triples between knowledge graph X1 and knowledge graph X2, calculate the correlation between knowledge graph X1 and knowledge graph X2, and if the correlation is greater than a preset correlation threshold, then use courseware template M as the target template. Several target templates are obtained, the corresponding tags of each target template are extracted, the suitability of each template is obtained from the current template selection platform, and a courseware template recommendation list is generated in descending order of suitability.

3. The method for generating safety education courseware based on a large model according to claim 2, characterized in that, Calculate the correlation between knowledge graph X1 and knowledge graph X2, including: A triple is a combination of keyword-relationship-keyword; the similarity between the first keyword, relation, and second keyword in any two triples is obtained, and the average is calculated to obtain the similarity between the two triples; By obtaining the number of triples N in knowledge graph X1 and the number of triples Q in knowledge graph X2, the correlation between knowledge graph X1 and knowledge graph X2 can be obtained. Where max() is used to find the maximum value, F n1 F n2 F nQ These are the similarities between the nth triplet in knowledge graph X1 and the 1st, 2nd, ..., Qth triplets in knowledge graph X2, respectively.

4. The method for generating safety education courseware based on a large model according to claim 2, characterized in that, The steps for calculating the conformity of a template include: extracting all tags corresponding to each target template, counting the occurrences of each tag, normalizing the occurrences, and using the normalization result as the distribution rate of each tag; obtaining all templates in the platform used by the current template, summing the distribution rates of each tag corresponding to a certain template, and using the sum as the conformity of that template.

5. The method for generating safety education courseware based on a large model according to claim 1, characterized in that, Determine the primary correlation for each safety incident case, including: Extract all keywords from the description of a specific explanatory section to create a first keyword set; obtain all keywords from the description of a specific safety accident case to create a second keyword set; extract all keywords from the first and second keyword sets and remove duplicates to construct a third keyword set. First and second word vectors with the same dimensions as the third keyword set are established, with the initial value of each dimension being 0. Each keyword in the third keyword set is matched and statistically analyzed in the first and second keyword sets to obtain the final first and second word vectors. The cosine similarity between the two word vectors is calculated as the first relevance of the safety accident cases in this explanation section.

6. The method for generating safety education courseware based on a large model according to claim 5, characterized in that, The final first and second word vectors are obtained by: if the i-th keyword in the third keyword set exists in the first keyword set, then the value of the i-th dimension in the first word vector is modified to 1; if the j-th keyword in the third keyword set exists in the second keyword set, then the value of the j-th dimension in the second word vector is modified to 1.

Images