Instruction information storage method, device, equipment and computer readable medium

Abstract

The embodiments of the present disclosure disclose instruction information storage methods, devices, equipment and computer-readable media. A specific implementation of the method includes: obtaining at least one target text data; for each target text data, executing the first generation step: generating specific prompt words; inputting the specific prompt words into a pre-trained large language model to generate a seed instruction information set; generating divergent prompt words; generating a divergent instruction information set; fusing the seed instruction information set and the divergent instruction information set; for each fused instruction information, executing the second generation step: determining feasibility information and innovative information; generating instruction evaluation information; removing fused instruction information that is not satisfied with the preset indicator evaluation conditions to obtain a candidate instruction information set; storing at least one candidate instruction information set. This implementation can quickly and efficiently realize diversified instruction information with divergent thinking.

Description

Technical Field

[0001] Embodiments of the present disclosure relate to the field of computer technology, and in particular to an instruction information storage method, apparatus, device, and computer-readable medium. Background Art

[0002] At present, with the continuous development of large language models, the application of large language models based on instruction data is becoming more and more extensive. For the generation of instruction data, the usual method is to obtain a fine-tuned instruction data set in the field through self-instruct technology.

[0003] However, when the above method is used to generate instruction data, the following technical problems often occur:

[0004] First, the divergent thinking of the instruction data set obtained through the self-instruct technology is limited, resulting in a low data set quality of the obtained instruction data set.

[0005] The second step is that the accuracy of instruction information is crucial. The measurement of instruction information in existing related technologies is not accurate enough. How to accurately measure the accuracy of instruction information is one of the current main research directions.

[0006] The above information disclosed in this Background section is only for enhancement of understanding of the background of the inventive concept and therefore it may contain information that does not form the prior art that is already known in this country to a person of ordinary skill in the art. Summary of the invention

[0007] The content of this disclosure is used to introduce concepts in a brief form, which will be described in detail in the detailed implementation section below. The content of this disclosure is not intended to identify the key features or essential features of the technical solution claimed for protection, nor is it intended to limit the scope of the technical solution claimed for protection.

[0008] Some embodiments of the present disclosure propose instruction information storage methods, devices, equipment, and computer-readable media to solve one or more of the technical problems mentioned in the above background technology section.

[0009] In a first aspect, some embodiments of the present disclosure provide an instruction information storage method, comprising: acquiring at least one target text data; for each target text data in the at least one target text data, executing the following first generation step: generating a specific prompt word corresponding to the target text data; inputting the specific prompt word into a pre-trained large language model to generate a seed instruction information set in a predetermined format; generating a divergent prompt word for the seed instruction information set; generating a divergent instruction information set corresponding to the seed instruction information set based on the divergent prompt word; fusing the seed instruction information set with the divergent instruction information set to generate a fused instruction information set; for each fused instruction information in the fused instruction information set, executing the following second generation step: determining feasibility information and innovation information corresponding to the fused instruction information; generating instruction evaluation information corresponding to the fused instruction information based on the feasibility information and the innovation information; removing the fused instruction information whose corresponding instruction evaluation information is not satisfied with the preset indicator evaluation condition from the fused instruction information set, and obtaining the removed instruction information set as a candidate instruction information set; and storing the obtained at least one candidate instruction information set.

[0010] In a second aspect, some embodiments of the present disclosure provide an instruction information storage device, including: an acquisition unit, configured to acquire at least one target text data;

[0011] The execution unit is configured to execute the following first generation step for each target text data in the at least one target text data: generate a specific prompt word corresponding to the target text data; input the specific prompt word into a pre-trained large language model to generate a seed instruction information set in a predetermined format; generate a divergent prompt word for the seed instruction information set; generate a divergent instruction information set corresponding to the seed instruction information set according to the divergent prompt word; fuse the seed instruction information set and the divergent instruction information set to generate a fused instruction information set; for each fused instruction information in the fused instruction information set, execute the following second generation step: determine the feasibility information and innovation information corresponding to the fused instruction information; generate instruction evaluation information corresponding to the fused instruction information according to the feasibility information and the innovation information; remove the fused instruction information whose corresponding instruction evaluation information is not satisfied with the preset indicator evaluation condition from the fused instruction information set, and obtain the removed instruction information set as the candidate instruction information set; the storage unit is configured to store the obtained at least one candidate instruction information set.

[0012] In a third aspect, some embodiments of the present disclosure provide an electronic device comprising: one or more processors; a storage device on which one or more programs are stored, and when the one or more programs are executed by the one or more processors, the one or more processors implement the method described in any implementation manner in the first aspect.

[0013] In a fourth aspect, some embodiments of the present disclosure provide a computer-readable medium having a computer program stored thereon, wherein when the program is executed by a processor, the method described in any implementation manner in the first aspect is implemented.

[0014] The above-mentioned various embodiments of the present disclosure have the following beneficial effects: through the instruction information storage method of some embodiments of the present disclosure, diversified instruction information with divergent thinking can be realized quickly and efficiently. Specifically, the reason for the low quality of the generated instruction information is that the divergent thinking of the instruction data set obtained by the self-instruct technology is limited, resulting in the low data set quality of the obtained instruction data set. Based on this, the instruction information storage method of some embodiments of the present disclosure, first, obtains at least one target text data as a text data source to facilitate the generation of subsequent instruction information. Then, for each target text data in the above-mentioned at least one target text data, the following first generation step is performed: the first step is to generate a specific prompt word corresponding to the above-mentioned target text data, so as to facilitate the subsequent quick and effective generation of a seed-level instruction information set for the target text data. The second step is to input the above-mentioned specific prompt word into a pre-trained large language model to accurately generate a seed instruction information set in a predetermined format. Here, the generated seed instruction information set is used as a seed-level basic instruction information set, which is convenient for the subsequent large language model to learn the information structure and instruction-related content of the seed instruction information to perform a divergent expansion of the instruction. The third step is to generate divergent prompt words for the above-mentioned seed instruction information set. Here, the generated divergent prompt words serve as inspiration information for the generation of divergent instruction information, which is convenient for the subsequent effective and quick generation of divergent instruction information sets. The fourth step is to quickly and effectively generate the divergent instruction information set corresponding to the above-mentioned seed instruction information set based on the above-mentioned divergent prompt words. Here, the generated divergent instruction information set is based on the seed instruction information set and is generated based on divergent thinking. Therefore, the divergent instruction information set represents the instruction information under various divergent thinking corresponding to the target text data. The fifth step is to fuse the above-mentioned seed instruction information set and the above-mentioned divergent instruction information set to generate a fused instruction information set, so as to facilitate the total evaluation of each instruction information. The sixth step is to perform the following second generation step for each fused instruction information in the above-mentioned fused instruction information set: Sub-step 1, determine the feasibility information and innovative information corresponding to the above-mentioned fused instruction information, so as to fully consider the instruction situation of each instruction information from the perspective of feasibility and innovation, so as to avoid the occurrence of low-quality instruction information. Sub-step 2, based on the above-mentioned feasibility information and the above-mentioned innovative information, the instruction evaluation information corresponding to the above-mentioned fused instruction information can be accurately generated. Here, the obtained instruction evaluation information can characterize the quality of the instruction information. In the seventh step, the fused instruction information whose corresponding instruction evaluation information is not satisfactory to the preset indicator evaluation condition is removed from the above fused instruction information set to obtain a high-quality removed instruction information set as a candidate instruction information set. Finally, the obtained at least one candidate instruction information set is stored for the convenience of the subsequent use of at least one candidate instruction information set.In summary, by generating divergent instruction information sets for divergent prompt words, the instruction information sets under diversified divergent thinking can be expanded to ensure the diversity of instruction information. In addition, through the judgment of feasibility and innovation, low-quality instruction information can be removed from the obtained diversified instruction information set to generate a high-quality instruction information set (i.e., at least one candidate instruction information set). BRIEF DESCRIPTION OF THE DRAWINGS

[0015] The above and other features, advantages and aspects of the embodiments of the present disclosure will become more apparent with reference to the following detailed description in conjunction with the accompanying drawings. Throughout the accompanying drawings, the same or similar reference numerals represent the same or similar elements. It should be understood that the drawings are schematic and that components and elements are not necessarily drawn to scale.

[0016] Figure 1 is a flow chart of some embodiments of the instruction information storage method according to the present disclosure;

[0017] Figure 2 is a schematic diagram of the structure of some embodiments of the instruction information storage device according to the present disclosure;

[0018] Figure 3 It is a schematic diagram of the structure of an electronic device suitable for implementing some embodiments of the present disclosure. DETAILED DESCRIPTION

[0019] Embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although certain embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that the present disclosure can be implemented in various forms and should not be construed as being limited to the embodiments set forth herein. On the contrary, these embodiments are provided to provide a more thorough and complete understanding of the present disclosure. It should be understood that the drawings and embodiments of the present disclosure are only for exemplary purposes and are not intended to limit the scope of protection of the present disclosure.

[0020] It should also be noted that, for ease of description, only the parts related to the invention are shown in the drawings. In the absence of conflict, the embodiments and features in the embodiments of the present disclosure can be combined with each other.

[0021] It should be noted that the concepts such as "first" and "second" mentioned in the present disclosure are only used to distinguish different devices, modules or units, and are not used to limit the order or interdependence of the functions performed by these devices, modules or units.

[0022] It should be noted that the modifications of "one" and "plurality" mentioned in the present disclosure are illustrative rather than restrictive, and those skilled in the art should understand that unless otherwise clearly indicated in the context, it should be understood as "one or more".

[0023] The names of the messages or information exchanged between multiple devices in the embodiments of the present disclosure are only used for illustrative purposes and are not used to limit the scope of these messages or information.

[0024] The present disclosure will be described in detail below with reference to the accompanying drawings and in conjunction with embodiments.

[0025] refer to Figure 1 , shows a process 100 of some embodiments of the instruction information storage method according to the present disclosure. The instruction information storage method comprises the following steps:

[0026] Step 101: Obtain at least one target text data.

[0027] In some embodiments, the execution subject (for example, an electronic device) of the above-mentioned instruction information storage method can obtain at least one target text data through a wired connection or a wireless connection. Among them, the target text data can be pre-selected text data for subsequent generation of an instruction information set. In practice, the target text data can be comments from artificial experts or text data discussed in a meeting. The instruction information can be instruction data corresponding to an artificial intelligence large language model. The instruction information is a data set consisting of (instruction, output) pairs. In practice, the specific form of the instruction information is a dictionary with a key in the form of {'instruction', 'input', 'output'}. For example, the instruction information can be {"instruction":"text11","input":"text22","output":"text33"}.

[0028] Step 102: for each target text data in the at least one target text data, perform the following first generation step:

[0029] Step 1021: Generate a specific prompt word corresponding to the target text data.

[0030] In some embodiments, the execution entity may generate specific prompt words corresponding to the target text data. The specific prompt words may be words that instruct the large language model to generate corresponding output results according to the requirements corresponding to the specific prompt words. In practice, the specific prompt words may be prompt words. The specific prompt words may generate prompt words for a seed instruction information set corresponding to the target text data. The seed instruction information set may be various instruction information after answering questions based on the specific prompt words for the target text data. That is, the seed instruction information set is various instruction information after the prompt words are input into the large language model. For example, the specific prompt words may be "read the content of the following meeting minutes (i.e., the target text data), extract the seed instruction data set of the instruction data set related to the domain xxx, which can be used to fine-tune the instructions related to the large language model and the domain xxx".

[0031] As an example, the execution subject may generate a specific prompt word corresponding to the target text data by using a preset prompt word template.

[0032] Step 1022: input the above-mentioned specific prompt words into a pre-trained large language model to generate a seed instruction information set in a predetermined format.

[0033] In some embodiments, the above-mentioned execution subject can input the above-mentioned specific prompt words into a pre-trained large language model to generate a seed instruction information set in a predetermined format. Among them, the predetermined format can be a pre-set instruction format. For example, the instruction format can be {'instruction', 'input', 'output'}. The large language model can be a pre-selected large language model. For example, the large language model can be an LLM model. The number of seed instruction information in the seed instruction information set is determined according to the provisions of the specific prompt words. For example, the seed instruction information set can be [{"instruction":"text1","input":"text2","output":"text3"}, {"instruction":"text4","input":"text5","output":"text6"}, {"instruction":"text7","input":"text8","output":"text9"}, {"instruction":"text10","input":"text11","output":"text12"}].

[0034] Step 1023, generating divergent prompt words for the above seed instruction information set.

[0035] In some embodiments, the execution subject may generate divergent prompt words for the seed instruction information set. The divergent prompt words may be prompt words that instruct the large language model to generate more instruction information by diverging thinking according to the format of the seed instruction information set. For example, the divergent prompt words may be the following examples that have been given as references, requiring you to generate 10 instructions, and the 'instruction' must involve the xxx field, and the generated instructions must involve solutions to related problems in this field, and at the same time meet the characteristics of yyy:

[0036] 1, {'instruction': text11, 'input': text12, 'output': text13};

[0037] 2.{'instruction':text21,'input':text22,'output':text23};

[0038] 3. {'instruction': text31, 'input': text32, 'output': text33}.

[0039] Among them, text{i}{j}1<=i<=3,1<=j<=3 represents the text corresponding to the instruction data set in a specific field.

[0040] The yyy text above could be: "When answering this question, please adopt a structured way of thinking. First, identify the core controversy or key points of the question. Next, list all potential assumptions and evaluate their validity. Then, use logical reasoning to connect these assumptions with your answer, ensuring that your argument has a clear causal chain. At the same time, identify and explain possible sources of bias, and take measures to reduce the impact of these biases on your answer. Finally, provide multiple angles to support your point of view and consider opposing views to enhance the comprehensiveness and depth of your answer."

[0041] As an example, the execution subject may generate divergent prompt words for the seed instruction information set through a divergent prompt word template.

[0042] In some optional implementations of some embodiments, the seed instruction information in the above-mentioned seed instruction information set includes: field information and at least one question and answer information group. The question and answer field corresponding to the above-mentioned at least one question and answer information group is the field corresponding to the above-mentioned field information. The question and answer information group includes: question information and reply information. Among them, the field information can be the field information of the content field corresponding to the instruction content corresponding to the seed instruction information. For example, the seed instruction information is instruction information for artificial intelligence, and the corresponding field information can be the field of artificial intelligence. The question information can be an inquiry question. The reply information can be information for replying to the inquiry question. For example, the question information can be "Is logistic regression a classification model or a regression model?" The corresponding reply information can be "Logistic regression can be a classification model."

[0043] Optionally, the above-mentioned generation of divergent prompt words for the above-mentioned seed instruction information set may include the following steps:

[0044] The first step is to determine at least one field information corresponding to the target text data. The at least one field information may be field information of various content fields involved in the text content corresponding to the target text data. For example, if the target text data is a science and technology magazine, the corresponding at least one field information includes: artificial intelligence field information, automation field information, and signal field information.

[0045] The second step is to obtain the pre-generated keyword knowledge graph and word-related knowledge graph set corresponding to each domain information. Among them, the keyword knowledge graph represents the corresponding relationship between each keyword, and each word-related knowledge graph has a corresponding keyword. Among them, each domain information has a corresponding keyword knowledge graph. Each node in the keyword knowledge graph is a keyword, and each edge represents the corresponding relationship between each keyword. The corresponding relationship can be but is not limited to: affiliation, synchronization, and proximity. For example, the domain information is artificial intelligence domain information, and the corresponding keyword knowledge graph is the knowledge graph of artificial intelligence keywords. For example, each keyword in the knowledge graph of artificial intelligence keywords may include: computer vision, natural language processing, audio processing, and multimodal processing. There are corresponding keywords in the word-related knowledge graph. That is, there is a one-to-one correspondence between the keywords in the keyword knowledge graph and the word-related knowledge graph in the word-related knowledge graph set. The word heat corresponding to each keyword in the keyword knowledge graph is greater than the set first threshold. The word heat can be reflected by the number of searches for the word, representing the intensity of the word being searched. The word heat corresponding to the word in the word-related knowledge graph is less than the first threshold and greater than the second threshold. The word-association knowledge graph is a knowledge graph that takes the keyword as the central word and has a direct or indirect relationship with the keyword. For example, if the keyword is computer vision, the words in the corresponding word-association knowledge graph include: residual model, random inactivation, pooling layer, FPN model, and attention mechanism. The edges in the word-association knowledge graph represent the corresponding relationship between the words.

[0046] In the third step, for each piece of domain information in the at least one piece of domain information, the following third generation step is performed:

[0047] Sub-step 1: determine the keyword knowledge graph and word association knowledge graph set corresponding to the above-mentioned domain information as the target keyword knowledge graph and target word association knowledge graph set. Among them, each keyword in the keyword knowledge graph has a corresponding relationship with the domain information. There is a one-to-one correspondence between the keywords in the keyword knowledge graph and the word association knowledge graph in the word association knowledge graph set.

[0048] As an example, first, the keyword knowledge graph corresponding to the domain information is obtained as the target keyword knowledge graph. Then, for each keyword in the target keyword knowledge graph, the word association knowledge graph corresponding to the keyword is determined as the target word association knowledge graph.

[0049] Sub-step 2: determining at least one seed instruction information in the seed instruction information set corresponding to the field information, wherein the field information corresponding to the seed instruction information in the at least one seed instruction information is the field information.

[0050] Sub-step 3: for each seed instruction information in the at least one seed instruction information, perform the following fourth generation step:

[0051] The first sub-step is to determine at least one question and answer information group corresponding to the seed instruction information as at least one target question and answer information group.

[0052] The second sub-step is to determine at least one target question information corresponding to the at least one target question and answer information group, wherein there is a one-to-one correspondence between the target question and answer information group in the at least one target question information group and the target question information in the at least one target question information.

[0053] The third sub-step is to determine the keyword group and related phrase corresponding to each target question information in the at least one target question information, and obtain at least one keyword group and at least one related phrase. The related phrase may be each word that has a word correlation relationship with the keyword in the keyword group. In practice, the word correlation relationship may be a word proximity relationship.

[0054] The fourth sub-step is to expand the at least one related phrase to generate at least one expanded phrase.

[0055] As an example, for each related word in at least one related word group, a similar word set corresponding to the related word is determined, and the obtained at least one similar word set is filled into at least one related word group to obtain at least one expanded word group.

[0056] The fifth sub-step is to determine the word clue information group corresponding to the keyword group for each keyword group and the corresponding expanded word group in at least one keyword group using the target keyword knowledge graph and the target word associated knowledge graph set. The word clue information may be a summary of a predetermined number of keywords to generate thinking clue information. That is, the word clue information is clue information for expanding thinking for a predetermined number of keywords. For example, the predetermined number of keywords may include: machine vision and multimodal processing. The corresponding word clue information may be "In the field of machine vision, multimodal target detection is achieved based on the attention mechanism."

[0057] As an example, first, at least one keyword that has a direct edge connection relationship with each keyword in the above keyword group is queried from the target keyword knowledge graph to obtain at least one keyword group. Then, the keyword group and at least one keyword group are keyword fused to generate a keyword set. Next, at least one keyword is extracted from the keyword set as a group of keywords to obtain at least one keyword subset. Among them, the number of words corresponding to a group of keywords is a predetermined number. For each keyword subset in at least one keyword subset, associated words that have a direct edge connection relationship with each keyword in the keyword subset are extracted from the corresponding target word associated knowledge graph subset to obtain an associated word set. Among them, the target word associated knowledge graph in the target word associated knowledge graph subset has a one-to-one correspondence with the keywords in the keyword subset. The associated word set and the keyword subset are word fused to obtain a fused word set. According to the fused word set, the corresponding word clue information is generated through the generative and adversarial neural network models. The obtained at least one word clue information is determined as a word clue information group.

[0058] The sixth sub-step is to determine the obtained at least one word clue information group as the word clue information set corresponding to the above-mentioned seed instruction information.

[0059] Sub-step 4, determining the full quantifier clue information set corresponding to the above-mentioned domain information. The full quantifier clue information set may be pre-set individual word clue information under the domain corresponding to the domain information. The full quantifier clue information set may be set by experts in the relevant domain for the domain information.

[0060] Sub-step 5, removing at least one word clue information set obtained from the above-mentioned full word clue information set to generate a word clue information set after removal.

[0061] Sub-step 6: generating at least one target word clue information corresponding to the above domain information based on the above removed word clue information set.

[0062] As an example, the execution subject may randomly extract a preset number of word clue information from the removed word clue information set to obtain at least one target word clue information.

[0063] Sub-step 7: generating a prompt word corresponding to the above-mentioned domain information according to the above-mentioned at least one target word clue information.

[0064] As an example, the execution entity may input at least one target word clue information into a prompt word filling template to generate a prompt word corresponding to the domain information.

[0065] The fourth step is to generate the divergent prompt word according to the at least one prompt word obtained.

[0066] As an example, the execution subject may input at least one prompt word into the divergent prompt word filling template to generate divergent prompt words.

[0067] In some optional implementations of some embodiments, the step of expanding the at least one related phrase to generate at least one expanded phrase may include the following steps:

[0068] The first step is to determine the keyword set corresponding to the keyword knowledge graph and the word set corresponding to the word association knowledge graph set. The keyword set corresponding to the keyword knowledge graph can be each keyword corresponding to each node in the keyword knowledge graph. The word set corresponding to the word association knowledge graph set can be the aggregated node set (i.e., word set) corresponding to each word association knowledge graph.

[0069] The second step is to fuse the above keyword set with the above word set to generate a fused word set.

[0070] The third step is to determine the word embedding vector corresponding to each fused word in the above fused word set as the first word embedding vector to generate a first word embedding vector set. The word embedding vector can represent the word semantic feature information corresponding to the fused word.

[0071] Step 4: for each related phrase in the at least one related phrase, perform the following adding steps:

[0072] Sub-step 1: determine the word embedding vector corresponding to each word in the above related word group as the second word embedding vector, and obtain the second word embedding vector group.

[0073] Sub-step 2: for each second word embedding vector in the second word embedding vector group, select from the first word embedding vector set at least one first word embedding vector whose vector distance with the second word embedding vector satisfies a preset distance condition. The preset distance condition may be that the cosine distance is less than a preset distance.

[0074] Sub-step 3: adding at least one word corresponding to the at least one first word embedding vector to the related word group to generate an expanded word group.

[0075] Optionally, the determining of the word embedding vector corresponding to each fused word in the fused word set as the first word embedding vector may include the following steps:

[0076] The first step is to determine the associated word set and associated edge set corresponding to the above fused word, wherein there is a one-to-one correspondence between the associated words in the above associated word set and the associated edges in the above associated edge set. The associated word can be a word directly connected to the fused word in the graph. The edge between the associated word and the fused word is the associated edge.

[0077] In the second step, for each associated word in the above associated word set, the following fifth generation step is performed:

[0078] Sub-step 1: determine the associated edge corresponding to the above associated words as the target associated edge.

[0079] Sub-step 2: Generate an associated word vector corresponding to the associated word and an associated edge vector corresponding to the target associated edge. The associated word vector can represent the semantic content features of the word corresponding to the associated word. The associated edge vector can represent the semantic content features of the associated relationship corresponding to the associated edge.

[0080] Sub-step 3, determining the initial fusion word vector corresponding to the fusion word, wherein the initial fusion word vector can represent the semantic content features corresponding to the fusion word.

[0081] Sub-step 4: concatenate the initial fused word vector, the associated word vector and the associated edge vector according to a predetermined vector order to generate a concatenated vector as the first word embedding vector.

[0082] Optionally, the above-mentioned use of the above-mentioned target keyword knowledge graph and the target word associated knowledge graph set to determine the word clue information group corresponding to the above-mentioned keyword group may include the following steps:

[0083] The first step is to annotate each keyword in the keyword group in the target keyword knowledge graph to generate an annotated keyword knowledge graph, wherein the keyword annotation may be a keyword highlight annotation.

[0084] The second step is to determine the keyword groups with keyword-term connection edges in the above-annotated keyword knowledge graph to obtain a keyword group set. Among them, the keyword-term connection edge can be an edge connecting two keywords in the annotated keyword knowledge graph. That is, there is an edge connecting two keywords in the keyword group.

[0085] The third step is to determine the word-related knowledge graph corresponding to each keyword in the above keyword group to obtain a word-related knowledge graph group.

[0086] The fourth step is to annotate each word-associated knowledge graph in the above-mentioned word-associated knowledge graph group corresponding to the expanded phrase to generate an annotated word-associated knowledge graph group.

[0087] Step 5: For each annotated word-related knowledge graph in the annotated word-related knowledge graph group, determine the annotated phrases in the annotated word-related knowledge graph that have a word connection edge, and obtain an annotated phrase set. The word connection edge may be an edge connecting two words in the annotated word-related knowledge graph. That is, two annotated words in the annotated phrase group have an edge connecting them.

[0088] Step 6: For each keyword in the above keyword group, perform the following sixth generation step:

[0089] Sub-step 1: determining at least one expansion word and at least one marked phrase corresponding to the above keyword as at least one first expansion word and at least one first marked phrase, respectively.

[0090] Sub-step 2: generating a first word clue information subset based on the above-mentioned keywords, the above-mentioned at least one first expanded word and the above-mentioned at least one first marked phrase.

[0091] As an example, the execution entity may input the keyword, at least one first expansion word, and the at least one first annotation phrase into a pre-trained Transformer model to generate a first word clue information subset. The Transformer model may be a neural network model that generates word clue information based on a multi-head attention mechanism. The Transformer model may be trained by a conventional training method.

[0092] Step 7: For each keyword group in the above keyword group set, perform the following seventh generation step:

[0093] Sub-step 1: determining at least one expanded word and at least one marked word group corresponding to the above keyword group as at least one second expanded word and at least one second marked word group, respectively.

[0094] Sub-step 2: Generate a second word clue information subset based on the keyword group, the at least one second expanded word and the at least one second marked word group. The specific implementation will not be repeated here, and reference may be made to the generation of the first word clue information subset.

[0095] In the eighth step, the above-mentioned word clue information group is generated according to the obtained first word clue information subset group and the obtained second word clue information subset.

[0096] As an example, the execution subject may combine the first word clue information subset group and the obtained second word clue information subset to generate a word clue information group.

[0097] Step 1024: Generate a divergent instruction information set corresponding to the seed instruction information set based on the divergent prompt words.

[0098] In some embodiments, the execution subject may generate a divergent instruction information set corresponding to the seed instruction information set according to the divergent prompt word, wherein the divergent instruction information may be an instruction information set after divergent thinking is expanded according to the seed instruction information under the prompt of the divergent prompt word.

[0099] As an example, the execution subject may input divergent prompt words into the large language model to generate a divergent instruction information set.

[0100] Step 1025, merging the seed instruction information set and the divergent instruction information set to generate a fused instruction information set.

[0101] In some embodiments, the execution entity may fuse the seed instruction information set and the divergent instruction information set to generate a fused instruction information set.

[0102] Step 1026: for each fusion instruction information in the fusion instruction information set, perform the following second generation step:

[0103] Step 10261, determine the feasibility information and innovation information corresponding to the above fusion instruction information.

[0104] In some embodiments, the execution subject may determine the feasibility information and innovation information corresponding to the fusion instruction information. The innovation information may represent the innovation degree of the instruction content corresponding to the fusion instruction information. The feasibility information may represent the feasibility degree of the instruction content corresponding to the fusion instruction information. Both the feasibility information and the innovation information may be information in the form of scores, and the higher the corresponding score, the greater the feasibility degree and the higher the innovation degree.

[0105] As an example, the execution entity may use the MCTS system framework to determine the feasibility information and innovation information corresponding to the fusion instruction information.

[0106] In some optional implementations of some embodiments, the above-mentioned execution entity can use a pre-trained text classification model to determine the feasibility information and innovative information corresponding to the above-mentioned fusion instruction information. Among them, the text classification model can be a two-category BERT model. Among them, in the two-category BERT model, the first token position (CLS position) output by the last layer of BERT is used as the representation of the sentence, followed by a fully connected layer for classification. This model is trained on data annotated by human experts. The text input of the classification model is the elements in the expanded seed instruction set (the dictionary is converted to text format), and the elements with high uncertainty are evaluated for feasibility through a large model. This can reduce the amount of annotation by human experts. The label is the feasibility annotated by human experts. 1 means feasible. 0 means infeasible.

[0107] Optionally, the text classification model may include: at least one feasibility information classification model and at least one innovative information classification model. In practice, there is a one-to-one correspondence between the feasibility information classification model in at least one feasibility information classification model and the innovative information classification model in at least one innovative information classification model. The feasibility information classification model may be a classification model for generating feasibility information. The feasibility information may be information in numerical form. The larger the numerical value, the higher the degree of feasibility. The innovative information classification model may be a classification model for generating innovative information. The innovative information may be information in numerical form. The larger the numerical value, the higher the degree of innovativeness. The feasibility information classification model and the corresponding innovative information classification model are trained synchronously. The network structures of the feasibility information classification models in at least one feasibility information classification model are different. The network structures of the innovative information classification models in at least one innovative information classification model are different. In practice, at least one feasibility information classification model may include: LSTM model, BERT model, Transformer model. Similarly, at least one innovative information classification model may include: LSTM model, BERT model, Transformer model.

[0108] Optionally, using a pre-trained text classification model, the feasibility information and innovation information corresponding to the above fusion instruction information are determined, including:

[0109] In the first step, for each feasibility information classification model in at least one feasibility information classification model, the following generation steps are performed:

[0110] Sub-step 1: input the above fusion instruction information into the feasibility information classification model to generate initial feasibility information.

[0111] Sub-step 2: determining an innovative information classification model among the at least one innovative information classification model corresponding to the feasibility information classification model as a target innovative information classification model.

[0112] Sub-step 3, inputting the above fusion instruction information into the target innovative information classification model to generate initial innovative information.

[0113] Sub-step 4, generating combined information of initial feasibility information and initial innovativeness information.

[0114] The second step is to determine the combination information in which the combination information represents the combination information corresponding to the initial feasibility information greater than the first value and the initial innovative information greater than the second value in the at least one combination information obtained, and obtain the combination information set.

[0115] In the third step, in response to determining that the number of information corresponding to the combined information set is greater than the third value, each combined information in the combined information set is subjected to weighted summation processing to generate weighted summation feasibility information and weighted summation feasibility innovation information, which are used as feasibility information and innovation information respectively.

[0116] The above-mentioned "optionally" content, as one of the inventive points of the present disclosure, solves the technical problem mentioned in the background technology that "the accuracy of the instruction information of the fused instruction information is crucial, the measurement of the instruction information of the existing related technology is not accurate enough, and how to accurately measure the accuracy of the fused instruction information is one of the current main research directions." Based on this, the present disclosure sets at least one feasibility information classification model and at least one innovative information classification model, and accurately generates feasibility information and innovative information through a combination of the feasibility information classification model and the innovative information classification model, so as to effectively measure the accuracy of the instruction information.

[0117] Step 10262, based on the feasibility information and the innovation information, generate instruction evaluation information corresponding to the fusion instruction information.

[0118] In some embodiments, the execution subject may generate instruction evaluation information corresponding to the fused instruction information based on the feasibility information and the innovation information, wherein the instruction evaluation information may be the quality of the instruction corresponding to the fused instruction information.

[0119] As an example, the execution entity may perform weighted sum processing on the feasibility information and the innovation information to generate instruction evaluation information.

[0120] Step 1027, removing the fused instruction information whose corresponding instruction evaluation information does not satisfy the preset indicator evaluation condition from the above fused instruction information set, and obtaining the instruction information set after removal as the candidate instruction information set.

[0121] In some embodiments, the execution subject may remove the fused instruction information whose corresponding instruction evaluation information is not satisfactory to the preset index evaluation condition from the fused instruction information set, and obtain the removed instruction information set as the candidate instruction information set. The preset index evaluation condition may be that the instruction evaluation information corresponding to the fused instruction information is greater than or equal to the preset evaluation score.

[0122] Step 103: store the obtained at least one candidate instruction information set.

[0123] In some embodiments, the execution entity may store the obtained at least one candidate instruction information set.

[0124] In some optional implementations of some embodiments, after step 103, the steps further include:

[0125] According to the at least one stored candidate instruction information set, perform one of the following operations:

[0126] Operation 1: retrain the large language model to generate a trained large language model as the large language model.

[0127] Operation 2, retraining the at least one sub-language model corresponding to at least one target text field in the large language model to generate a trained large language model as the large language model, wherein the at least one target text field is at least one text field corresponding to the at least one target text data. There is a one-to-one correspondence between the target text field in the at least one target text field and the sub-language model in the at least one sub-language model. For example, if the target text field is a target detection field, the corresponding sub-language model may be a target detection model.

[0128] The above-mentioned various embodiments of the present disclosure have the following beneficial effects: through the instruction information storage method of some embodiments of the present disclosure, diversified instruction information with divergent thinking can be realized quickly and efficiently. Specifically, the reason for the low quality of the generated instruction information is that the divergent thinking of the instruction data set obtained by the self-instruct technology is limited, resulting in the low data set quality of the obtained instruction data set. Based on this, the instruction information storage method of some embodiments of the present disclosure, first, obtains at least one target text data as a text data source to facilitate the generation of subsequent instruction information. Then, for each target text data in the above-mentioned at least one target text data, the following first generation step is performed: the first step is to generate a specific prompt word corresponding to the above-mentioned target text data, so as to facilitate the subsequent quick and effective generation of a seed-level instruction information set for the target text data. The second step is to input the above-mentioned specific prompt word into a pre-trained large language model to accurately generate a seed instruction information set in a predetermined format. Here, the generated seed instruction information set is used as a seed-level basic instruction information set, which is convenient for the subsequent large language model to learn the information structure and instruction-related content of the seed instruction information to perform a divergent expansion of the instruction. The third step is to generate divergent prompt words for the above-mentioned seed instruction information set. Here, the generated divergent prompt words serve as inspiration information for the generation of divergent instruction information, which is convenient for the subsequent effective and quick generation of divergent instruction information sets. The fourth step is to quickly and effectively generate the divergent instruction information set corresponding to the above-mentioned seed instruction information set based on the above-mentioned divergent prompt words. Here, the generated divergent instruction information set is based on the seed instruction information set and is generated based on divergent thinking. Therefore, the divergent instruction information set represents the instruction information under various divergent thinking corresponding to the target text data. The fifth step is to fuse the above-mentioned seed instruction information set and the above-mentioned divergent instruction information set to generate a fused instruction information set, so as to facilitate the total evaluation of each instruction information. The sixth step is to perform the following second generation step for each fused instruction information in the above-mentioned fused instruction information set: Sub-step 1, determine the feasibility information and innovative information corresponding to the above-mentioned fused instruction information, so as to fully consider the instruction situation of each instruction information from the perspective of feasibility and innovation, so as to avoid the occurrence of low-quality instruction information. Sub-step 2, based on the above-mentioned feasibility information and the above-mentioned innovative information, the instruction evaluation information corresponding to the above-mentioned fused instruction information can be accurately generated. Here, the obtained instruction evaluation information can characterize the quality of the instruction information. In the seventh step, the fused instruction information whose corresponding instruction evaluation information is not satisfactory to the preset indicator evaluation condition is removed from the above fused instruction information set to obtain a high-quality removed instruction information set as a candidate instruction information set. Finally, the obtained at least one candidate instruction information set is stored for the convenience of the subsequent use of at least one candidate instruction information set.In summary, by generating divergent instruction information sets for divergent prompt words, the instruction information sets under diversified divergent thinking can be expanded to ensure the diversity of instruction information. In addition, through the judgment of feasibility and innovation, low-quality instruction information can be removed from the obtained diversified instruction information set to generate a high-quality instruction information set (i.e., at least one candidate instruction information set).

[0129] Further references Figure 2 As an implementation of the methods shown in the above figures, the present disclosure provides some embodiments of an instruction information storage device, and these device embodiments are Figure 1 Corresponding to the method embodiments shown, the instruction information storage device can be specifically applied to various electronic devices.

[0130] like Figure 2 As shown, an instruction information storage device 200 includes: an acquisition unit 201, an execution unit 202 and a storage unit 203. Among them, the acquisition unit 201 is configured to acquire at least one target text data; the execution unit 202 is configured to perform the following first generation step for each target text data in the above at least one target text data: generate a specific prompt word corresponding to the above target text data; input the above specific prompt word into a pre-trained large language model to generate a seed instruction information set in a predetermined format; generate a divergent prompt word for the above seed instruction information set; generate a divergent instruction information set corresponding to the above seed instruction information set according to the above divergent prompt word; fuse the above seed instruction information set and the above divergent instruction information set to generate a fused instruction information set; for each fused instruction information in the above fused instruction information set, perform the following second generation step: determine the feasibility information and innovation information corresponding to the above fused instruction information; generate instruction evaluation information corresponding to the above fused instruction information according to the above feasibility information and the above innovation information; remove the fused instruction information whose corresponding instruction evaluation information is not satisfied with the preset indicator evaluation condition from the above fused instruction information set, and obtain the removed instruction information set as the candidate instruction information set; the storage unit 203 is configured to store the obtained at least one candidate instruction information set.

[0131] It can be understood that the units recorded in the instruction information storage device 200 and the reference Figure 1 Therefore, the operations, features and beneficial effects described above for the method are also applicable to the instruction information storage device 200 and the units contained therein, and will not be described in detail here.

[0132] Reference below Figure 3 , which shows a structural schematic diagram of an electronic device (eg, an electronic device) 300 suitable for implementing some embodiments of the present disclosure. Figure 3 The electronic device shown is only an example and should not bring any limitation to the functions and scope of use of the embodiments of the present disclosure.

[0133] like Figure 3 As shown, the electronic device 300 may include a processing device (e.g., a central processing unit, a graphics processing unit, etc.) 301, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 302 or a program loaded from a storage device 308 to a random access memory (RAM) 303. In the RAM 303, various programs and data required for the operation of the electronic device 300 are also stored. The processing device 301, the ROM 302, and the RAM 303 are connected to each other via a bus 304. An input / output (I / O) interface 305 is also connected to the bus 304.

[0134] Typically, the following devices may be connected to the I / O interface 305: an input device 306 including, for example, a touch screen, a touch pad, a keyboard, a mouse, a camera, a microphone, an accelerometer, a gyroscope, etc.; an output device 307 including, for example, a liquid crystal display (LCD), a speaker, a vibrator, etc.; a storage device 308 including, for example, a magnetic tape, a hard disk, etc.; and a communication device 309. The communication device 309 may allow the electronic device 300 to communicate with other devices wirelessly or by wire to exchange data. Figure 3 The electronic device 300 is shown with various devices, but it should be understood that it is not required to implement or possess all the devices shown. More or fewer devices may be implemented or possessed instead. Figure 3 Each block shown in the figure may represent one device, or may represent multiple devices as required.

[0135] In particular, according to some embodiments of the present disclosure, the process described above with reference to the flowchart can be implemented as a computer software program. For example, some embodiments of the present disclosure include a computer program product, which includes a computer program carried on a computer-readable medium, and the computer program includes a program code for executing the method shown in the flowchart. In some such embodiments, the computer program can be downloaded and installed from the network through the communication device 309, or installed from the storage device 308, or installed from the ROM 302. When the computer program is executed by the processing device 301, the above-mentioned functions defined in the method of some embodiments of the present disclosure are executed.

[0136] It should be noted that the computer-readable medium in some embodiments of the present disclosure may be a computer-readable signal medium or a computer-readable storage medium or any combination of the two. The computer-readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or device, or any combination of the above. More specific examples of computer-readable storage media may include, but are not limited to: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In some embodiments of the present disclosure, the computer-readable storage medium may be any tangible medium containing or storing a program, which may be used by or in combination with an instruction execution system, device or device. In some embodiments of the present disclosure, the computer-readable signal medium may include a data signal propagated in a baseband or as part of a carrier wave, in which a computer-readable program code is carried. Such propagated data signals may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. Computer readable signal media may also be any computer readable medium other than computer readable storage media, which may send, propagate or transmit a program for use by or in conjunction with an instruction execution system, apparatus or device. The program code contained on the computer readable medium may be transmitted using any appropriate medium, including but not limited to: wires, optical cables, RF (radio frequency), etc., or any suitable combination of the above.

[0137] In some embodiments, the client and the server may communicate using any currently known or future developed network protocol such as HTTP (HyperText Transfer Protocol), and may be interconnected with any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include a local area network ("LAN"), a wide area network ("WAN"), an internet (e.g., the Internet), and a peer-to-peer network (e.g., an ad hoc peer-to-peer network), as well as any currently known or future developed network.

[0138] The computer-readable medium may be included in the electronic device, or may exist independently without being incorporated into the electronic device. The computer-readable medium carries one or more programs. When the one or more programs are executed by the electronic device, the electronic device: obtains at least one target text data; for each target text data in the at least one target text data, performs the following first generation step: generates a specific prompt word corresponding to the target text data; inputs the specific prompt word into a pre-trained large language model to generate a seed instruction information set in a predetermined format; generates a divergent prompt word for the seed instruction information set; generates a divergent instruction information set corresponding to the seed instruction information set according to the divergent prompt word; fuses the seed instruction information set with the divergent instruction information set to generate a fused instruction information set; for each fused instruction information in the fused instruction information set, performs the following second generation step: determines the feasibility information and innovation information corresponding to the fused instruction information; generates instruction evaluation information corresponding to the fused instruction information according to the feasibility information and the innovation information; removes the fused instruction information whose corresponding instruction evaluation information is not satisfied with the preset indicator evaluation condition from the fused instruction information set to obtain the removed instruction information set as the candidate instruction information set; and stores the obtained at least one candidate instruction information set.

[0139] Computer program code for performing the operations of some embodiments of the present disclosure may be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, C++, and conventional procedural programming languages ​​such as "C" or similar programming languages. The program code may be executed entirely on the user's computer, partially on the user's computer, as a separate software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In the case of a remote computer, the remote computer may be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0140] The flow chart and block diagram in the accompanying drawings illustrate the possible architecture, function and operation of the system, method and computer program product according to various embodiments of the present disclosure. In this regard, each square box in the flow chart or block diagram can represent a module, a program segment or a part of a code, and the module, the program segment or a part of the code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some implementations as replacements, the functions marked in the square box can also occur in a sequence different from that marked in the accompanying drawings. For example, two square boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each square box in the block diagram and / or flow chart, and the combination of the square boxes in the block diagram and / or flow chart can be implemented with a dedicated hardware-based system that performs a specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.

[0141] The units described in some embodiments of the present disclosure may be implemented by software or hardware. The units described may also be provided in a processor, for example, may be described as: a processor includes an acquisition unit, an execution unit, and a storage unit. The names of these units do not, in some cases, constitute limitations on the units themselves, for example, the acquisition unit may also be described as a "unit for acquiring at least one target text data".

[0142] The functions described above herein may be performed at least in part by one or more hardware logic components. For example, without limitation, exemplary types of hardware logic components that may be used include: field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems on chips (SOCs), complex programmable logic devices (CPLDs), and the like.

[0143] The above descriptions are only some preferred embodiments of the present disclosure and an explanation of the technical principles used. Those skilled in the art should understand that the scope of the invention involved in the embodiments of the present disclosure is not limited to the technical solutions formed by a specific combination of the above technical features, but should also cover other technical solutions formed by any combination of the above technical features or their equivalent features without departing from the above inventive concept. For example, the above features are replaced with (but not limited to) technical features with similar functions disclosed in the embodiments of the present disclosure.

Claims

1. A method for storing instruction information, comprising: Acquire at least one target text data; For each target text data in the at least one target text data, the following first generating step is performed: Generating a specific prompt word corresponding to the target text data; Input the specific prompt word into a pre-trained large language model to generate a seed instruction information set in a predetermined format, wherein the seed instruction information in the seed instruction information set includes: field information and at least one question and answer information group, the question and answer field corresponding to the at least one question and answer information group is the field corresponding to the field information, and the question and answer information group includes: question information and reply information; Generate divergent prompt words for the seed instruction information set, wherein: The generating of divergent prompt words for the seed instruction information set includes: Determine at least one field information corresponding to the target text data; Obtain the pre-generated keyword knowledge graph and word-related knowledge graph set corresponding to each field information, wherein the keyword knowledge graph represents the corresponding relationship between each keyword, and each word-related knowledge graph has a corresponding keyword; For each piece of domain information in the at least one domain information, the following third generating step is performed: Determine the keyword knowledge graph and word association knowledge graph set corresponding to the domain information as the target keyword knowledge graph and target word association knowledge graph set; Determine at least one seed instruction information in the seed instruction information set corresponding to the field information; For each seed instruction information of the at least one seed instruction information, the following fourth generating step is performed: Determine at least one question and answer information group corresponding to the seed instruction information as at least one target question and answer information group; Determine at least one target question information corresponding to the at least one target question and answer information group; Determine a keyword group and a related word group corresponding to each target question information in the at least one target question information, and obtain at least one keyword group and at least one related word group; performing word expansion on the at least one related word group to generate at least one expanded word group; For each keyword group and the corresponding expanded word group in at least one keyword group, using the target keyword knowledge graph and the target word associated knowledge graph set, determine the word clue information group corresponding to the keyword group; Determining the obtained at least one word clue information group as the word clue information set corresponding to the seed instruction information; Determine a full quantifier clue information set corresponding to the domain information; removing at least one word clue information set obtained from the full word clue information set to generate a word clue information set after removal; generating at least one target word clue information corresponding to the domain information according to the removed word clue information set; generating a prompt word corresponding to the domain information according to the at least one target word clue information; Generate the divergent prompt word according to the at least one prompt word obtained; Generating a divergent instruction information set corresponding to the seed instruction information set according to the divergent prompt word; Fusing the seed instruction information set and the divergent instruction information set to generate a fused instruction information set; For each fusion instruction information in the fusion instruction information set, the following second generation step is performed: Determine feasibility information and innovation information corresponding to the fusion instruction information, wherein the innovation information represents the innovation degree of the instruction content corresponding to the fusion instruction information, and the feasibility information represents the feasibility degree of the instruction content corresponding to the fusion instruction information; Generating instruction evaluation information corresponding to the fusion instruction information according to the feasibility information and the innovation information; Remove the fused instruction information whose corresponding instruction evaluation information is not satisfactory to the preset indicator evaluation condition from the fused instruction information set, and obtain the instruction information set after removal as the candidate instruction information set; The obtained at least one candidate instruction information set is stored.

2. The method according to claim 1, wherein: The method further comprises: According to the at least one stored candidate instruction information set, perform one of the following operations: Retraining the large language model to generate a trained large language model as the large language model; Retraining the at least one sub-language model corresponding to at least one target text field in the large language model to generate a trained large language model as the large language model, wherein the at least one target text field is at least one text field corresponding to the at least one target text data.

3. The method according to claim 2, wherein: The step of expanding the at least one related phrase to generate at least one expanded phrase comprises: Determine a keyword set corresponding to the keyword knowledge graph and a word set corresponding to the word association knowledge graph set; fusing the keyword set with the word set to generate a fused word set; Determine a word embedding vector corresponding to each fused word in the fused word set as a first word embedding vector to generate a first word embedding vector set; For each related word group in the at least one related word group, the following adding steps are performed: Determine a word embedding vector corresponding to each word in the related word group as a second word embedding vector, and obtain a second word embedding vector group; For each second word embedding vector in the second word embedding vector group, select from the first word embedding vector set at least one first word embedding vector whose vector distance with the second word embedding vector satisfies a preset distance condition; At least one word corresponding to the at least one first word embedding vector is added to the related word group to generate an expanded word group.

4. The method according to claim 3, wherein: The step of determining the word embedding vector corresponding to each fused word in the fused word set as the first word embedding vector includes: Determine an associated word set and an associated edge set corresponding to the fused word, wherein there is a one-to-one correspondence between the associated words in the associated word set and the associated edges in the associated edge set; For each associated word in the associated word set, the following fifth generation step is performed: Determine the associated edge corresponding to the associated word as the target associated edge; Generate an associated word vector corresponding to the associated word and an associated edge vector corresponding to the target associated edge; Determine an initial fused word vector corresponding to the fused word; The initial fused word vector, the associated word vector and the associated edge vector are concatenated according to a predetermined vector order to generate a concatenated vector as the first word embedding vector.

5. The method according to claim 1, wherein: The step of using the target keyword knowledge graph and the target word associated knowledge graph set to determine the keyword group corresponding word clue information group includes: Annotate each keyword in the keyword group in the target keyword knowledge graph to generate an annotated keyword knowledge graph; Determine the keyword groups with keyword-term connection edges in the annotated keyword knowledge graph to obtain a keyword group set; Determine a word-associated knowledge graph corresponding to each keyword in the keyword group to obtain a word-associated knowledge graph group; Annotate each word-associated knowledge graph of the corresponding expanded phrase in the word-associated knowledge graph group to generate an annotated word-associated knowledge graph group; For each annotated word-associated knowledge graph in the annotated word-associated knowledge graph group, determining an annotated phrase having a word-connected edge in the annotated word-associated knowledge graph to obtain an annotated phrase set; For each keyword in the keyword group, the following sixth generation step is performed: Determining at least one expanded word and at least one marked phrase corresponding to the keyword as at least one first expanded word and at least one first marked phrase, respectively; generating a first word clue information subset according to the keyword, the at least one first expanded word and the at least one first marked word group; For each keyword group in the keyword group set, the following seventh generation step is performed: Determining at least one expanded word and at least one marked word group corresponding to the keyword group as at least one second expanded word and at least one second marked word group, respectively; generating a second word clue information subset according to the keyword group, the at least one second expanded word and the at least one second marked word group; The word clue information group is generated according to the obtained first word clue information subset group and the obtained second word clue information subset.

6. The method according to claim 1, wherein: The determining of the feasibility information and innovative information corresponding to the fusion instruction information includes: The feasibility information and innovative information corresponding to the fusion instruction information are determined by using a pre-trained text classification model.

7. A command information storage device, comprising: An acquisition unit, configured to acquire at least one target text data; The execution unit is configured to perform the following first generation step for each target text data in the at least one target text data: generate a specific prompt word corresponding to the target text data; input the specific prompt word into a pre-trained large language model to generate a seed instruction information set in a predetermined format, wherein the seed instruction information in the seed instruction information set includes: domain information and at least one question and answer information group, the question and answer domain corresponding to the at least one question and answer information group is the domain corresponding to the domain information, and the question and answer information group includes: question information and reply information; generate divergent prompt words for the seed instruction information set, wherein the generation of the divergent prompt words for the seed instruction information set includes: determining the at least one domain information corresponding to the target text data; obtaining a pre-generated keyword knowledge graph and a word association knowledge graph set corresponding to each domain information, wherein the keyword knowledge graph represents the correspondence between each keyword, and each word association knowledge graph has a corresponding keyword; for each domain information in the at least one domain information, executing the following third generation step: determining the keyword knowledge graph and the word association knowledge graph set corresponding to the domain information as the target keyword knowledge graph and the target word association knowledge graph set; determining at least one seed instruction information in the seed instruction information set corresponding to the domain information; for each seed instruction information in the at least one seed instruction information, executing the following fourth generation step The method comprises the following steps: determining at least one question-and-answer information group corresponding to the seed instruction information as at least one target question-and-answer information group; determining at least one target question information corresponding to the at least one target question-and-answer information group; determining a keyword group and a related phrase corresponding to each target question information in the at least one target question information, and obtaining at least one keyword group and at least one related phrase; performing word expansion on the at least one related phrase to generate at least one expanded phrase; for each keyword group and the corresponding expanded phrase in the at least one keyword group, using the target keyword knowledge graph and the target word associated knowledge graph set, determining a word clue information group corresponding to the keyword group; and determining the obtained at least one word clue information group as The word clue information set corresponding to the seed instruction information; determining the full word clue information set corresponding to the domain information; removing at least one word clue information set obtained from the full word clue information set to generate a word clue information set after removal; generating at least one target word clue information corresponding to the domain information according to the removed word clue information set; generating a prompt word corresponding to the domain information according to the at least one target word clue information; generating the divergent prompt word according to the obtained at least one prompt word; generating a divergent instruction information set corresponding to the seed instruction information set according to the divergent prompt word; fusing the seed instruction information set and the divergent instruction information set to generate a fused instruction information set;For each fusion instruction information in the fusion instruction information set, the following second generation step is performed: determining feasibility information and innovation information corresponding to the fusion instruction information; generating instruction evaluation information corresponding to the fusion instruction information according to the feasibility information and the innovation information; removing fusion instruction information whose corresponding instruction evaluation information is not satisfactory to a preset indicator evaluation condition from the fusion instruction information set, and obtaining a post-removal instruction information set as a candidate instruction information set; The storage unit is configured to store the obtained at least one candidate instruction information set.

8. An electronic device comprising: one or more processors; a storage device having one or more programs stored thereon, When the one or more programs are executed by the one or more processors, the one or more processors implement the method according to any one of claims 1 to 6.

9. A computer readable medium having a computer program stored thereon, wherein: When the program is executed by a processor, the method according to any one of claims 1 to 6 is implemented.

Images