Key intelligence demand intention understanding method and system based on large model

Abstract

The invention discloses a key intelligence demand intention understanding method based on a large model. The method comprises the following steps: constructing a key intelligence demand intention understanding large model; the large model comprises an interaction instruction understanding algorithm, a combat scheme understanding algorithm, a context prompt learning technology, a heterogeneous intelligence association fusion algorithm and a multi-modal intelligence understanding algorithm; establishing automatic intelligence analysis and understanding by combining large-model efficient parameter fine tuning, retrieval enhancement generation and cue word technology; intention understanding is carried out on key intelligence demands based on a large model; the decision-making module is used for decision-making based on intelligence, and the decision-making based on intelligence comprises associated commander key information demand generation, task decision point mining and appointed traction branches and subsequent plans. The invention further discloses a system, electronic equipment and a computer readable storage medium, based on a military big language model, efficient parameter fine tuning is utilized by the big model, an iterative intelligence vector database and a flexible intelligence task cue word template are implemented, and the intelligent processing capacity of intelligence data is remarkably improved.

Description

Technical Field

[0001] The present invention relates to the fields of big models, intelligence technology, computer simulation and modeling simulation technology, and in particular to a method and system for understanding key intelligence demand intentions based on big models. Background Art

[0002] Artificial intelligence has gradually entered the era of large-scale pre-trained models since the transformer in 2017. BERT uses the encoder's "cloze" method to train the model. By fine-tuning the model, it can achieve good results in specific tasks and has attracted attention from the industry. GPT uses the decoder to predict the next word to train the model, and its characteristic lies in its "generation" ability. On November 30, 2022, OpenAI introduced technologies such as instruction fine-tuning, supervised fine-tuning, and reinforcement learning based on human feedback (RLHF) into the model, and finally launched a new conversational general artificial intelligence tool - ChatGPT, which shows a high level of human-computer interaction. With the large amount of knowledge stored in the model, ChatGPT has achieved good application results in human-computer question and answer, text summary generation, machine translation, classification, code generation and other aspects. Artificial intelligence technologies based on large language models are gradually being applied to all aspects of scientific research, production and life. These advanced technologies enable machines to learn and reason autonomously. Their application ranges from situational awareness to decision support, demonstrating strong productivity and changing working methods and research paradigms in various fields including the military and government organizations.

[0003] The big language model has two core advantages, namely the ability to understand and execute complex instructions and the ability to decompose and plan complex tasks. The ability to understand and execute complex instructions is reflected in the fact that when we give the big model detailed instructions and clearly express the task constraints or specifications, the big model can complete the task as required by the instructions. This ability to be faithful to the instruction requirements is highly related to the big model's contextual generation ability. Given reasonable prompts, and the richer and more detailed the prompts are, the higher the quality of content that the big model can generate. This contextual generation ability essentially reflects an ability to model the world and has nothing to do with the way humans perceive the world. The ability to decompose and plan complex tasks is another advantage of the big model. It can decompose complex tasks into multiple steps and reasonably plan the execution order of tasks. This provides an important opportunity for vertical applications, enabling the big model to work with traditional information systems, efficiently coordinating many systems such as databases, knowledge bases, office automation systems, and code bases in traditional IT systems, and completing complex decision-making tasks that traditional intelligent systems have been unable to handle in the past, thereby improving the intelligence level of the entire information system.

[0004] The key technologies in the existing technical solutions involving large language models include:

[0005] 1. Efficient parameter fine-tuning technology for large language models

[0006] The working principle of the Large Language Model (LLM) is to predict the next word based on the input text through each node in the model neural network, and complete the answer in an autoregressive manner. The neural network nodes that determine the final answer in the entire autoregressive generation process of the model only account for a very small part of the total nodes. The neural network contains many dense layers to perform matrix multiplication operations, and the weight matrices in these layers are usually full rank. When fine-tuned for a specific task, the LLM has a lower "intrinsic dimension (intrinsic rank)" and can still be effectively learned even when randomly projected to a smaller subspace. Therefore, when constructing a military large language model in the intelligence mission scenario, this patent also selects LoRA (Low-Rank Adaptation), a low-rank adaptation-based technology, as one of the technical approaches for fine-tuning. This technology can achieve a good effect by only training a small number of parameters when using a large model to adapt to downstream tasks, and promote the model to have higher perception, cognition and decision-making intelligence in multi-source and multi-modal intelligence mission scenarios. LoRA has been widely used in open source LLMs (such as LLaMA and BLOOM) to achieve efficient parameter fine-tuning. By freezing the pre-trained model weights and injecting the trainable rank decomposition matrix into each layer of the Transformer architecture, LoRA greatly reduces the number of trainable parameters for downstream tasks and effectively improves the fine-tuning efficiency of the pre-trained model on downstream tasks.

[0007] like Figure 1 The technical architecture of LoRA is shown in the figure. The left part of the rounded rectangle in the figure is the original PLM (Pre-trained Language Model). LoRA adds two structures A and B next to it. The parameters of these two structures are initialized to Gaussian distribution and 0 respectively, realizing a dimensionality reduction and dimensionality increase operation to simulate the intrinsic rank (Intrinsic Rank). Initially, A is initialized with a random Gaussian distribution, and B is initialized with a 0 matrix to ensure that the bypass matrix is ​​still a 0 matrix at the beginning of training. During training, the parameters of PLM are fixed, and only the dimensionality reduction matrix A and the dimensionality increase matrix B are trained. The input and output dimensions of the model remain unchanged, and the parameters of BA and PLM are superimposed at output.

[0008] Assuming that you want to fine-tune a pre-trained language model (such as LLaMA) in the downstream task, you need to update the pre-trained model parameters. The formula is as follows:

[0009] W0+ΔW

[0010] W0 is the parameter initialized by the pre-trained model, and ΔW is the parameter that needs to be updated. If it is full parameter fine-tuning, its parameter amount is W0. It can be seen that the cost of full parameter fine-tuning of large language models is very high. For LORA, only ΔW needs to be fine-tuned.

[0011] Specifically, assuming that the pre-trained matrix is Its update can be expressed as:

[0012]

[0013] where rank r< <min(d,k)。

[0014] During the training process of LoRA, W0 is fixed, and only A and B are training parameters. In the forward process, W0 and ΔW are multiplied by the same input x and finally added:

[0015] h=W0x+ΔWx=W0x+BAx

[0016] Full Fine-Tuning can be seen as a special case of LoRA (when r is equal to k). During the inference process, LoRA also introduces almost no additional inference delay, and only needs to calculate W = W0 + ΔW. The combination of LoRA and Transformer is also very simple, only adding a bypass in the calculation of QKV Attention.

[0017] In the above fine-tuning process, only the parameters of A and B are updated, and the parameters of the pre-trained model are fixed. When performing subsequent reasoning, the idea of ​​re-parametrization is used to merge AB and W, so that no additional calculations are introduced during inference. And for different downstream tasks, only AB needs to be retrained based on the pre-trained model, which can also speed up the training pace of large models.

[0018] In summary, LoRA does not require the cumulative gradient updates of the weight matrix to have full rank during the adaptation process. When LoRA is applied to all weight matrices and all deviations are trained, setting the rank r of LoRA to the rank of the pre-trained weight matrix can roughly restore the expressiveness of full fine-tuning. In other words, as the number of trainable parameters increases, training LoRA roughly converges to training the original model; in addition, LoRA has no additional reasoning delay: when a specific individual agent is deployed in the joint training platform, W=W0+BA can be explicitly calculated and stored, and reasoning can be performed normally. When it is necessary to switch to another single-soldier Agent for deployment, W0 can be restored by subtracting BA and then adding a different B′A′. This is a fast operation that requires little memory overhead. Compared with the model with fine-tuning on the structural parameters, LoRA does not introduce any additional delays in the reasoning process. This ability to switch tasks at a lower cost during deployment can help this patent create more customized single-soldier entity models that can be switched in real time on machines that store pre-trained weights in VRAM; finally, LoRA can reduce memory and storage resource consumption. For large Transformers trained with Adam, if r< <d model ,LoRA reduces VRAM usage by 2 / 3 because it does not need to store the optimizer state of the fixed pre-trained parameters W0, and can use fewer GPUs for large model training.

[0019] 2. Prompt word engineering

[0020] The goal of prompt words is to obtain better answers from the big language model. The process is simply to repeatedly try and continuously iterate better prompt words, speech structures and external knowledge to ask questions to meet the actual and specific needs of users. Prompt engineering is a series of processes to find more suitable answers in the big language model through more suitable language organization based on understanding the capabilities and limitations of the big language model. Prompt words can be a question, a text description, or even a text model with a bunch of parameters. We interact with the big model through prompt words, and the model will also generate corresponding text, pictures, or videos and voices based on the information provided by the prompt words. Prompt words can often contain roles, instructions, context information, questions, output requirements, etc. Among them, the instruction part describes the task we need the big model to do. On the one hand, context information can provide some additional information to the model so that it can better execute your instructions, and on the other hand, it can define the information scope of the model. Questions, as the name suggests, are questions that need to be answered by the model. Output requirements can be the format, number of items or style that the model needs to follow in answering.

[0021] MattNight proposed a relatively complete prompt word methodology framework CSISPE on ChatGPT-Free-Prompt-List. In the CRISE framework, CR stands for Capacity and Role, which means the role and ability that the big model is expected to play. I stands for Insight, which is background information and context. S stands for Statement, which describes the task that the model is expected to complete. P stands for Personality, which is used to describe the style or method in which you want the big model to answer you. E stands for Experiment, which requires the model to provide multiple answers. The following examples of prompt words based on the CSISP framework are shown in Table 1:

[0022] Table 1 Examples of prompt words in the CSISP framework

[0023]

[0024] Specifically, instructions are mainly in the following forms:

[0025] Zero-shot prompts use the big model to do the task directly without providing reference examples. As the most common way to use prompts, zero-shot prompts make full use of the knowledge learned by the big model. They are the most flexible and universal, and are also one of the important scenarios for evaluating the capabilities of the big model. At the same time, zero-shot prompts rely on pre-trained language models, which may be limited and biased by training data.

[0026] Few-shot prompts enable interaction between people and large models by giving a small amount of prompt text, making the model output more accurate.

[0027] Role Prompt: By playing a role-playing game with the big model, the big model can imagine itself as an expert in a certain field to achieve better task results.

[0028] Chain-of-thought prompts are commonly used in phrase deduction tasks to solve difficult reasoning problems. The zero-shot chain-of-thought prompt is to put a sentence "Let's think step by step" at the end of the question. Using chain-of-thought prompts can make the model's output more coherent and logical. However, we need to note that if an error occurs in a certain step of the chain of thought, the error will gradually accumulate, resulting in the generated result not being consistent with expectations.

[0029] Multimodal Prompt, a multimodal prompt word, allows for better interaction with large models by simultaneously inputting multimodal information such as text and images.

[0030] In summary, the prior art has the following defects:

[0031] 1. Traditional intelligence processing system

[0032] Traditional intelligence processing systems rely mainly on manual operations, using keyword search techniques and information retrieval algorithms. These systems typically extract information that matches specific keywords by searching documents, reports, and other records in databases. Human intelligence analysts are responsible for screening, organizing, and writing intelligence reports.

[0033] The technical defects are its low efficiency and slow manual operation, especially when faced with large amounts of intelligence data, the processing efficiency is low, and it is difficult to respond quickly to battlefield needs; in addition, keyword searches can easily miss information that does not contain keywords but is still important, resulting in incomplete intelligence; the system is highly dependent on manual operation and requires a lot of manpower, which increases operating and labor costs; and the system cannot automatically understand and analyze complex intelligence needs, has a low level of intelligence, and the quality and efficiency of intelligence analysis are limited by manual levels.

[0034] 2. Rule-based intelligence analysis system

[0035] Rule-based intelligence analysis systems use predefined rules and logic for intelligence processing. These rules are usually designed by experts and include simple decision trees, logistic regression and other algorithms to automatically classify and analyze intelligence data.

[0036] The technical flaw is that it lacks flexibility. The system can only operate according to predefined rules and has difficulty adapting to the rapidly changing battlefield environment and complex and changing intelligence needs. It is also difficult to maintain. The rules and logic need to be constantly updated and maintained to cope with new intelligence situations and changing environments, which increases the complexity and maintenance costs of the system. In addition, the system performs poorly and lacks the ability to respond to unforeseen new situations and non-standardized data.

[0037] 3. Machine learning and data mining technology

[0038] Using machine learning and data mining algorithms, valuable intelligence information can be extracted from large-scale data. These technologies include classification, clustering, association rule mining, etc., which can automatically analyze and predict intelligence trends.

[0039] The technical defect is that it is highly data-dependent. Model training requires a large amount of high-quality labeled data, which is costly and time-consuming to obtain and annotate. In addition, when processing unseen data or new scenarios, machine learning models are often not as effective as those in the training data set, resulting in insufficient generalization capabilities. In addition, some complex machine learning models (such as deep learning) are difficult to explain their internal analysis processes and results, and commanders may not trust or be unable to fully utilize these intelligence analysis results.

[0040] 4. Natural Language Processing (NLP) Technology

[0041] Automatic intelligence report generation based on natural language processing technologies, including word vectors, topic models, and deep learning, which are used to understand and analyze text data.

[0042] The technical defect is that its understanding ability is limited. Traditional NLP models have limited understanding of complex contexts and implicit information, and may not be able to fully and accurately analyze and interpret intelligence information. In addition, NLP systems lack the ability to process real-time changing information and dynamic situations, and cannot adjust the focus and direction of analysis in real time. In addition, NLP models require a lot of pre-training and fine-tuning, which is costly, and have weak adaptability in specific fields, requiring constant adjustment to meet specific intelligence needs. Summary of the invention

[0043] The purpose of the present invention is to construct a key intelligence demand intention understanding technology for task decision points based on a large model in view of the defects of the prior art, and to design a key intelligence demand intention understanding method and system based on a large model, including: designing an automated intelligence analysis and understanding scheme that combines efficient parameter fine-tuning, retrieval enhancement generation and prompt word technology of a large model, refining the efficient parameter fine-tuning steps and designing basic prompt word templates for different stages of intelligence tasks, so that under the empowerment of multiple intelligent technologies such as a large language model, by fine-tuning the existing military large language model in the intelligence task scenario, supplemented by the relevant ideas of prompt engineering, the model can efficiently and high-quality meet the decision-makers, analysts and other decision-makers' needs for key intelligence, build a transparent battlefield from the perspective of intelligence, improve the intelligence level of intelligence data retrieval, processing and application, ensure the timeliness and relevance of intelligence, and optimize intelligence task planning through customized prompt word templates and fine-tuning data sets, enhance the close integration of intelligence and decision-making, achieve efficient information filtering and sorting, and ultimately improve the intelligence processing efficiency and decision-making quality in military operations.

[0044] The first aspect of the present invention is to provide a method for understanding key intelligence requirements based on a large model, comprising:

[0045] S1, constructing a large model for understanding the intention of key intelligence needs; wherein the large model for understanding the intention of key intelligence needs includes an interactive instruction understanding algorithm, an operational plan understanding algorithm, a contextual prompt learning technology, a heterogeneous intelligence association fusion algorithm, and a multimodal intelligence understanding algorithm; the large model for understanding the intention of key intelligence needs is established based on the automated intelligence analysis and understanding of the joint large model efficient parameter fine-tuning, retrieval enhancement generation, and prompt word technology;

[0046] S2, understanding the intent of key intelligence requirements based on the key intelligence requirement intention understanding big model; wherein the intelligence-based decision-making includes associating commanders' key information requirements generation, mining mission decision points, and specifying traction branches and subsequent plans.

[0047] Preferably, the S1 comprises:

[0048] S11, fine-tuning the basic military large language model enabled by chatglm through initial intelligence data to obtain a fine-tuned model;

[0049] S12, based on the fine-tuning model and based on the prompt word engineering technology, construct a prompt word template for a specific field, so that the fine-tuning model can meet the needs of different intelligence tasks.

[0050] Preferably, the S11 includes:

[0051] S111, construct three types of data sets, namely, intelligence instruction fine-tuning data set, intelligence vector database and preset prompt word template library; wherein:

[0052] The intelligence instruction fine-tuning dataset is used for fine-tuning the military large language model to enhance the model's ability to understand and respond to specific types of military tasks. It is a high-quality fine-tuning dataset designed and generated based on specific military intelligence tasks, and contains task-related questions and answers, as well as specific scenarios, terms, protocols and other elements.

[0053] The intelligence vector database integrates historical intelligence mission data, including past inquiries, reports, analysis results, etc., and integrates new data generated by experts into the database to ensure the timeliness and relevance of information;

[0054] The preset prompt word template library stores multiple templates designed in advance for different types of intelligence tasks to ensure the accuracy and efficiency of input;

[0055] S112, for a new original question, first perform keyword extraction to identify the core elements and themes of the question, and based on these keywords, select the most appropriate template from a preset prompt word template library; then, based on the intelligence instruction fine-tuning data set, use efficient parameter fine-tuning technology to fine-tune the basic military large language model to form a military large language model adapted to the intelligence field;

[0056] S113, fine-tuning large military models adapted to the intelligence field based on efficient parameter fine-tuning methods.

[0057] Preferably, the S113 includes:

[0058] (1) Collect and filter data to obtain intelligence data from different data sources and data modalities; including:

[0059] Collecting raw intelligence data: Collecting raw data from one or more data sources including open source intelligence, human intelligence, signals intelligence, geospatial intelligence, and technical intelligence;

[0060] Screening of raw intelligence data: performing data cleaning, data verification, data classification and labeling, data analysis, data fusion, and confidentiality and ethical processing on the raw intelligence data, and collecting and screening key information from the multi-source and multi-modal intelligence data as the data basis for the training of the large model and decision support;

[0061] (2) constructing multiple data models based on the integration of the filtered intelligence data and the characteristics and requirements of each intelligence; wherein the multiple data models correspond to each intelligence category one by one and are used to manage and analyze complex intelligence data; the data models include:

[0062] Entity-relationship model (ERM) and text analysis model for open source intelligence;

[0063] Graph database models and time-driven models for human intelligence;

[0064] Time series models and network traffic analysis models for signals intelligence;

[0065] Geographic Information System (GIS) model and multidimensional data model (OLAP) for geospatial intelligence;

[0066] Database models and association rule learning models for technical intelligence;

[0067] (3) constructing different types of data sets based on the constructed data model; wherein the data sets include text understanding data sets, intelligence-related data sets, decision point mining data sets, and auxiliary task planning data sets;

[0068] The text understanding datasets include: interactive instruction understanding datasets and multimodal intelligence understanding datasets;

[0069] The intelligence-related data sets include: intelligence report data sets, operational intelligence requirements understanding data sets, and intelligence mission planning and decomposition data sets;

[0070] The decision point mining data set includes: a decision point case data set and a scenario simulation data set;

[0071] The auxiliary mission planning data set includes: command and control communication data set and tactics and strategy guidance data set;

[0072] (4) constructing an intelligence vector data set and an expected intelligence instruction fine-tuning data set based on the data set, wherein the intelligence vector data set is updated in real time and is stored in real time whenever there is the latest real intelligence data; the expected intelligence instruction fine-tuning data set is stored in a data architecture that complies with the instruction fine-tuning specification after being screened, checked and supplemented by military experts, and is used to fine-tune the basic military large language model. The expected intelligence instruction fine-tuning data set is updated periodically according to the fine-tuning requirements;

[0073] (5) Fine-tune the military large model adapted to the intelligence field based on multi-modal instruction fine-tuning technology and efficient parameter fine-tuning technology;

[0074] The multimodal instruction fine-tuning technology is used to generate an intelligence instruction fine-tuning dataset based on prompt templates and experts through a large language, wherein the intelligence instruction fine-tuning dataset includes a seed instruction set and a corresponding input and output instance set to generate a large amount of new instruction data and corresponding input and output instances, and then use the generated instruction dataset to fine-tune the instructions of downstream tasks on the large model;

[0075] The efficient parameter fine-tuning technology includes setting a model and a word segmenter and selecting a suitable parameter quantization model according to the current fine-tuning data volume.

[0076] Preferably, the step S12, based on the fine-tuning model and based on the prompt word engineering technology, constructs a prompt word template for a specific field so that the fine-tuning model meets the needs of different intelligence tasks, including: constructing the prompt word template for different intelligence task scenarios; the different intelligence task scenarios include:

[0077] (1) Mission scenarios for intelligence data retrieval, processing, and application;

[0078] (2) Task decision point mining process task scenario;

[0079] (3) mission scenarios for the key information requirements generation process at mission decision points; and

[0080] (4) The branches driven by the decision point and the task scenarios for formulating subsequent plans.

[0081] Preferably, the prompt word template adopts a generalized prompt word template as a basis, and an effective prompt contains one or more of the following five elements: role, instruction, context information, input data and output indication; a basic prompt word template is constructed based on historical data, and specific prompt word templates are designed for typical different intelligence task scenarios based on the basic prompt word template.

[0082] Preferably, for the intelligence data retrieval, processing and application process task scenario, S12 includes:

[0083] S121, determine the mapping relationship between each stage in combat intelligence preparation and the task scenario of the intelligence data retrieval, processing and application process; wherein the processing of the intelligence data includes intelligence processing and intelligence analysis; for intelligence data retrieval, it corresponds to the intelligence plan formulation and guidance stage and the intelligence collection stage; for intelligence processing, it corresponds to the intelligence processing and utilization stage and the intelligence analysis and production stage; for intelligence analysis, it corresponds to the intelligence distribution and fusion stage and the intelligence evaluation and feedback stage; for intelligence application, it corresponds to the traditional document service and intelligence product provision stage and the content and knowledge output stage based on understanding user intentions and related concepts, including document information provision, professional analysis, processing and handling of intelligence products;

[0084] S122, designing the prompt word template based on the mapping relationship;

[0085] For the task decision point mining process task scenario, S12 includes:

[0086] S121', determining the application scenario of the task scenario of the task decision point mining process;

[0087] S122', designing the prompt word template based on the application scenario of the task scenario in the task decision point mining process and the prompt word technology of LLM;

[0088] For the task scenario of the key information requirement generation process for the task decision point, S12 includes:

[0089] S121”, determine the application scenario of the task scenario of the key information requirement generation process for the task decision point;

[0090] S122”, designing the prompt word template based on the application scenario of the task scenario of the key information requirement generation process for the task decision point and natural language processing and deep learning technology;

[0091] For the branch driven by the decision point and the subsequent plan formulation task scenario, S12 includes:

[0092] S121”′, determining the application scenario of the branch driven by the decision point and the formulation task scenario of the subsequent plan;

[0093] S122'', designing the prompt word template based on the branches driven by the decision point and the application scenario of the subsequent plan formulation task scenario.

[0094] The second aspect of the present invention provides a key intelligence requirement intention understanding system based on a large model, which is used to implement the method described in the first aspect, including:

[0095] A model building module (101) is used to build a key intelligence demand intention understanding big model; wherein the key intelligence demand intention understanding big model includes an interactive instruction understanding algorithm, an operational plan understanding algorithm, a contextual prompt learning technology, a heterogeneous intelligence association fusion algorithm and a multimodal intelligence understanding algorithm; the key intelligence demand intention understanding big model is established based on the automated intelligence analysis and understanding of the joint big model efficient parameter fine-tuning, retrieval enhancement generation and prompt word technology;

[0096] A key intelligence requirement intention understanding module (102) is used to understand the key intelligence requirement intention based on the key intelligence requirement intention understanding large model; wherein the intelligence-based decision-making includes associating commander key information requirement generation, mining task decision points, and specifying traction branches and subsequent plans.

[0097] A third aspect of the present invention provides an electronic device, comprising a processor and a memory, wherein the memory stores a plurality of instructions, and the processor is configured to read the instructions and execute the method described in the first aspect.

[0098] A fourth aspect of the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium stores a plurality of instructions, and the plurality of instructions can be read by a processor and execute the method described in the first aspect.

[0099] The core invention of the present invention is:

[0100] Method for constructing an intelligence vector database and an intelligence command fine-tuning dataset. This method designs a real-time updated intelligence vector database that collects historical intelligence mission data, including inquiries, reports, analysis results, etc., and retrieves information quickly and efficiently through advanced data processing technologies (such as vectorization and indexing). In addition, an expert-generated intelligence command fine-tuning dataset is designed and constructed for multimodal command fine-tuning of basic military large-scale language models. With the support of this solution, the timeliness and relevance of intelligence data are ensured, the understanding and response capabilities of large models for specific military tasks are improved, and the retrieval, processing and application processes of intelligence data are optimized. An intelligence vector database and an intelligence command fine-tuning dataset are constructed to ensure real-time information updates and quality control. In the task reasoning stage, the model dynamically adjusts the focus and direction of analysis by quickly retrieving samples in the intelligence vector database as context resources.

[0101] A chatglm-based large model fine-tuning method. Use the intelligence command fine-tuning dataset to fine-tune the basic military large language model for specific intelligence tasks to enhance the model's understanding and response accuracy for specific types of military tasks. Ultimately, improve the model's performance in different mission scenarios, enhance the model's ability to adjust real-time data of generated content, and promote mission decision point mining and key intelligence demand generation. Based on ChatGLM technology, perform efficient parameter fine-tuning on the military large language model to enhance the model's understanding and response capabilities for specific military tasks.

[0102] Intelligence mission planning prompt template driven by mission decision points. Through the flexible construction of prompt word templates, customized information filtering and sorting options are provided, and examples and other information are organically combined to form prompt inputs for large language models. The model output is adapted to the preferences and needs of different commanders, and the information display and analysis methods can be automatically adjusted according to the current mission and environment, improving the efficiency of intelligence writing and human-computer interaction. The preset prompt word template library realizes the customization and systematization of intelligence mission planning, ensures the accuracy and efficiency of input, automatically generates high-quality intelligence that meets the current mission requirements, and provides efficient information filtering and sorting options to enhance the close integration of intelligence and decision-making.

[0103] Beneficial effects of the method and system of the present invention:

[0104] The present invention is based on a military large language model, and utilizes efficient parameter fine-tuning of the large model, implementation of an iterative intelligence vector database, and flexible intelligence task prompt word templates to significantly enhance the intelligent processing capabilities of intelligence data. Compared with traditional systems that rely on manual keyword search and writing, the present invention can more quickly identify and extract key information in the following aspects, and improve the efficiency and accuracy of intelligence retrieval, integration, and analysis.

[0105] Real-time update and efficient retrieval. By building an intelligence vector database, the present invention ensures that intelligence information can be updated in real time to adapt to the rapidly changing military environment. Compared with the static and lagging data processing in traditional intelligence systems, the present invention can dynamically adjust the focus and direction of analysis and provide more timely and relevant intelligence support.

[0106] Customized intelligence task planning. The present invention realizes the customization and systematization of intelligence task planning through a preset prompt word template library. The templates in the prompt word template library are designed in advance according to different types of intelligence tasks to ensure the accuracy and efficiency of the input, so that the system can generate the optimal intelligence plan according to specific task requirements and enhance the close integration of intelligence and decision-making.

[0107] Efficient parameter fine-tuning. The ChatGLM technology is used to efficiently fine-tune the parameters of the military large language model, which enhances the model's ability to understand and respond to specific military tasks. Compared with traditional models, this invention can more accurately generate intelligence that adapts to different mission scenarios and meet the personalized needs of commanders.

[0108] Automated intelligence generation. The present invention realizes the automation of intelligence generation by constructing an intelligence instruction fine-tuning data set and an intelligence vector database, combined with efficient parameter fine-tuning technology. Compared with traditional intelligence systems, the present invention can quickly retrieve and analyze historical data, organically combine relevant information to form input, automatically generate high-quality intelligence that meets the current task requirements, and significantly improve intelligence processing efficiency.

[0109] Flexible adjustment of analysis focus. In the inference stage, the large language model can adjust the analysis focus and direction according to real-time data by quickly retrieving samples in the intelligence vector database as context resources. Compared with traditional systems with fixed analysis modes, the present invention has stronger adaptability and flexibility, and can provide optimal intelligence analysis and decision support according to changes in tasks and environments. BRIEF DESCRIPTION OF THE DRAWINGS

[0110] In order to more clearly illustrate the specific embodiments of the present invention or the technical solutions in the related technologies, the drawings required for use in the specific embodiments or the related technical descriptions will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0111] Figure 1 The schematic diagram of the technical architecture of LoRA provided according to the prior art.

[0112] Figure 2 A schematic diagram of a process architecture provided according to an embodiment of the present invention;

[0113] Figure 3 A schematic diagram of a technical architecture provided according to an embodiment of the present invention;

[0114] Figure 4 A schematic diagram of the LoRA weight merging process flow according to an embodiment of the present invention;

[0115] Figure 5 A flow chart of a method for understanding key intelligence requirements and intentions based on a large model according to an embodiment of the present invention;

[0116] Figure 6 This is a diagram of the architecture of a key intelligence requirement intention understanding system based on a big model according to an embodiment of the present invention;

[0117] Figure 7 A structural diagram of an electronic device provided according to an embodiment of the present invention. DETAILED DESCRIPTION

[0118] The technical solution of the present invention will be described clearly and completely below in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0119] In the description of the present invention, it should be noted that the terms "center", "upper", "lower", "left", "right", "vertical", "horizontal", "inner", "outer", etc., indicating the orientation or positional relationship, are based on the orientation or positional relationship shown in the drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as limiting the present invention. In addition, the terms "first", "second", and "third" are used for descriptive purposes only, and cannot be understood as indicating or implying relative importance.

[0120] In the description of the present invention, it should be noted that, unless otherwise clearly specified and limited, the terms "installed", "connected", and "connected" should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection, or it can be indirectly connected through an intermediate medium, or it can be the internal communication of two components. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0121] like Figure 2-3As shown, the advantage of the large language model is that it is a "generalist". Many of its potential capabilities need us to be familiar with, discover, and even "awaken" in our daily work, so that it can play a greater role alone or in combination with intelligence-specific tools. In the context of generalists, the purpose of this patent is to fine-tune the basic military large language model enabled by ChatGLM through initial intelligence data. On the basis of the fine-tuned model, based on the prompt word engineering technology, a prompt word template for a specific field is constructed so that the model meets the needs of different intelligence tasks. In the early stage of intelligence preparation, based on the intelligence vector database, the intelligence data retrieval, processing and application process is optimized; in the mission planning stage, the mining of mission decision points is promoted; when the commander reaches the mission decision point, the generation of the commander's key intelligence needs is associated; in the decision-making process, facing the mission decision point, the formulation of branches and subsequent plans is driven.

[0122] This embodiment aims to solve multiple technical problems existing in the traditional military intelligence processing and application process, including:

[0123] (1) First, traditional intelligence systems rely on manual keyword searching and writing, which results in low intelligence retrieval efficiency, easy omission of key information, and a response speed that cannot meet rapidly changing military needs.

[0124] (2) Second, existing intelligence analysis lacks intelligence and is unable to fully understand and respond to complex military mission requirements, resulting in a disconnect between intelligence and decision-making.

[0125] (3) Third, the timeliness and accuracy of intelligence data processing are insufficient, making it difficult to provide timely and effective support in a dynamic environment.

[0126] (4) Fourth, intelligence mission planning lacks systematization and automation, and is unable to generate the key intelligence required by commanders based on mission decision points, affecting the scientific nature and effectiveness of decision-making.

[0127] Embodiment 1

[0128] like Figure 5 As shown, this embodiment provides a method for understanding key intelligence demand intention based on a large model, including:

[0129] S1, constructs a large model for understanding the intention of key intelligence needs; wherein the large model for understanding the intention of key intelligence needs includes an interactive instruction understanding algorithm, an operational plan understanding algorithm, a contextual prompt learning technology, a heterogeneous intelligence association fusion algorithm, and a multimodal intelligence understanding algorithm; in this embodiment, the S1 includes an automated intelligence analysis and understanding solution that combines efficient parameter fine-tuning of the large model, retrieval enhancement generation, and prompt word technology, refines the efficient parameter fine-tuning steps, and designs basic prompt word templates for different stages of intelligence tasks; by constructing a "customized preset prompt word template library (updated on demand) + intelligence vector database (updated in real time) + fine-tuning large model", rapid response and in-depth understanding of complex intelligence needs in different mission scenarios.

[0130] As a preferred embodiment, the S1 includes:

[0131] S11, fine-tuning the basic military large language model enabled by chatglm through initial intelligence data to obtain a fine-tuned model;

[0132] S12, based on the fine-tuning model and based on the prompt word engineering technology, construct a prompt word template for a specific field, so that the fine-tuning model can meet the needs of different intelligence tasks.

[0133] As a preferred implementation, the S11 includes:

[0134] S111, construct three types of data sets, namely, intelligence instruction fine-tuning data set, intelligence vector database and preset prompt word template library; wherein:

[0135] The intelligence instruction fine-tuning dataset is used for fine-tuning of the military large language model to enhance the model's understanding ability and response accuracy for specific types of military tasks. It is a high-quality fine-tuning dataset designed and generated based on specific military intelligence tasks (such as combat plan generation and intelligence mission planning), and contains task-related questions and answers, as well as specific scenarios, terms, protocols and other elements.

[0136] The intelligence vector database integrates historical intelligence mission data, including past inquiries, reports, analysis results, etc., and integrates new data generated by experts into the database to ensure the timeliness and relevance of information;

[0137] The preset prompt word template library stores a plurality of templates designed in advance for different types of intelligence tasks to ensure the accuracy and efficiency of input.

[0138] S112, for a new original question, first perform keyword extraction to identify the core elements and themes of the question, and based on these keywords, select the most appropriate template from the preset prompt word template library; then, based on the intelligence instruction fine-tuning data set, use efficient parameter fine-tuning technology to fine-tune the basic military large language model to form a military large language model adapted to the intelligence field.

[0139] The application methods of these three types of data sets in real application scenarios include:

[0140] In the use of a certain mission scenario, first retrieve historical cases and reports similar to the original question in the intelligence vector database, select highly relevant historical information as a reference, and use it as part of the input to improve the model's understanding of the current question and the relevance of the answer. In addition, retrieve and obtain the prompt word template with the highest degree of adaptability to the current mission goal in the preset prompt word template library, and effectively connect the original question, task examples and other important prompt words to form the final input. Finally, based on this carefully prepared input, the military large language model will generate the corresponding answer. This answer is designed to meet the specific intelligence needs raised by the original question. The rest of the manual will describe in detail the model fine-tuning and prompt word template design scheme in the intelligence mission scenario.

[0141] S113, fine-tuning the military large model adapted to the intelligence field;

[0142] The scale of pre-trained language models (PLMs) represented by ChatGPT is becoming larger and larger. Limited by computing power and training costs, full fine-tuning is expensive, and it is difficult to tune large models in all specific intelligence task scenarios. The parameter-efficient fine-tuning (PEFT) method is proposed to solve these two problems. PEFT can make PLM efficiently adapt to various downstream application tasks without fine-tuning all parameters of the pre-trained model. It can fine-tune a small number or additional model parameters and fix most of the pre-training parameters, which greatly reduces computing and storage costs. At the same time, the most advanced PEFT technology can also achieve performance comparable to full fine-tuning.

[0143] The embodiment of this patent adopts efficient parameter fine-tuning methods such as LoRA to conduct preliminary fine-tuning tests without changing the internal parameters of the model, confirming that the constructed data set is valid, and then synchronously fine-tuning all parameter instructions to update the internal parameters of the model to form specific capabilities. The detailed construction plan is described as follows.

[0144] As a preferred implementation, the S113 includes:

[0145] (1) Collect and filter data to obtain intelligence data with different data sources and data modalities;

[0146] Collecting data includes: the quality and quantity of data determines the model's "ability to make decisions from data". In order to effectively fine-tune, the data needs to be properly structured. When building a large military language model in intelligence-related mission scenarios, collecting multi-source and multi-modal data is a key step. First, it is necessary to collect intelligence data from different data sources and different data modalities. This patent embodiment is designed to collect from the sources shown in Table 2 below.

[0147] Table 2 Main data sources

[0148]

[0149] Screening data includes: cleaning, verifying, classifying and labeling the intelligence data, collecting and screening key information from multi-source and multi-modal intelligence, and providing a high-quality and highly relevant data foundation for subsequent model training and decision support. Such data processing not only enhances the accuracy and reliability of intelligence, but also improves the ability to understand complex combat environments.

[0150] The processing flow is shown in Table 3 below.

[0151] Table 3 Data processing flow

[0152]

[0153]

[0154] (2) constructing a plurality of data models based on the integration of the filtered intelligence data and the characteristics and requirements of each intelligence; wherein the plurality of data models correspond one-to-one to each intelligence category and are used to manage and analyze complex intelligence data;

[0155] For different types of intelligence, the construction of the data structure model needs to consider the characteristics and requirements of each type of intelligence. Based on the data source in step (1), the designed and constructed data model is shown in Table 4 below.

[0156] Table 4: Examples of partially designed data models

[0157]

[0158] The patented design of the above model integrates data from different intelligence sources. By tailoring the data structure model for each intelligence category, it can more effectively manage and analyze complex military intelligence data, support more accurate and in-depth intelligence work, and ultimately provide a comprehensive analytical perspective. For example, combining OSINT and SIGINT information to analyze a specific event or trend. In addition, the data model should be designed to be easily expanded and updated to adapt to the addition of new intelligence and changes in analysis requirements.

[0159] (3) Based on the constructed data model, it is expected to construct data sets of different categories, and the detailed classification is expected to be shown in Table 5 below:

[0160] Table 5 Expected datasets to be constructed

[0161]

[0162]

[0163] (4) Based on the above four categories, a total of nine data sets, an intelligence vector data set and an expected intelligence instruction fine-tuning data set are constructed respectively. The intelligence vector data set is updated in real time, and whenever there is the latest real intelligence data, it is stored in real time; the expected intelligence instruction fine-tuning data set is screened, checked and supplemented by military experts and stored in a data architecture that meets the instruction fine-tuning specifications, which is used to fine-tune the instructions of the basic military large language model. The expected intelligence instruction fine-tuning data set is periodically updated according to the fine-tuning needs.

[0164] (5) Fine-tune the military large model adapted to the intelligence field based on multi-modal instruction fine-tuning technology and efficient parameter fine-tuning technology;

[0165] In this embodiment:

[0166] Multimodal instruction fine-tuning technology: A large language fine-tuning dataset of intelligence instructions generated based on prompt templates and experts, which contains a seed instruction set and the corresponding input and output instance sets to generate a large amount of new instruction data and corresponding input and output instances, and then use the generated instruction dataset to fine-tune the instructions of the large model for downstream tasks. Compared with knowledge distillation, the construction of this instruction dataset has two significant differences: ① Knowledge distillation uses the knowledge of the large model to guide the training of the small model, and this method uses the large model itself to guide the fine-tuning of the large model in downstream tasks; ② Knowledge distillation can only extract knowledge within the field learned by the large model, while the instruction data and corresponding input and output instances generated by this method are mostly outside the data distribution of the pre-training stage.

[0167] Efficient parameter fine-tuning technology: After the hardware and software environment is built, set the model and word segmenter, and select the appropriate parameter quantization model based on the current amount of fine-tuning data. If you want to complete the fine-tuning process in a shorter time and with less memory, give priority to the 4-bit quantization provided by BitsAndBytesConfig. Then load and process the data, and the dataset is loaded based on the file format specified by the data_path parameter. Load the dataset from a JSON or JSON Lines file (specified as a file path by the data_path parameter). Then, use tokenize to preprocess the data. Concatenate the input and output text data and encode them as tag IDs, and append a sequence end marker. Then prepare the encoded data as input to the model.

[0168] Before calling LoRA for LLM fine-tuning, hyperparameters need to be set in advance. In terms of technical implementation, this patent design calls the LoraConfig method provided by the PeFT framework to configure LoRA parameters for subsequent fine-tuning. Table 6 below is a brief explanation of the key hyperparameters and an explanation of the selection rules. In the subsequent repeated fine-tuning work, it is necessary to adjust the following hyperparameters, especially lora_alpha and lora_rank, by observing the performance and resource consumption changes of the model to find the best settings for specific tasks.

[0169] Table 6 Hyperparameters Introduction

[0170]

[0171]

[0172] During fine-tuning, the smaller weight matrices A and B are separated. But once training is complete, the weights can be merged into an identical new weight matrix. Although LoRA is significantly smaller and faster to train, latency issues may be encountered during subsequent inference due to loading the base model and LoRA model separately. In order to eliminate latency, this patent selects the merge_and_unload function to merge the adapter weights with the base model, effectively using the newly merged model as an independent model. At the same time, multiple LoRA models can also be merged, allowing the Base Model to have the ability to handle multiple tasks at the same time.

[0173] The merging process is as follows Figure 4 As shown, W merged =W+BA replaces the original W. Multiple Lora models can be merged through the merge_and_unlaod method. By merging CKPTs of different types of tasks, the original model can have the ability of multiple downstream tasks at the same time without affecting the reasoning efficiency. Call the save_pretrained method to save the model.

[0174] As a preferred implementation, the S12, based on the fine-tuning model and based on the prompt word engineering technology, constructs a prompt word template for a specific field so that the fine-tuning model meets the needs of different intelligence tasks, including: constructing the prompt word template for different intelligence task scenarios; the different intelligence task scenarios include:

[0175] (1) Mission scenarios for intelligence data retrieval, processing, and application;

[0176] (2) Task decision point mining process task scenario;

[0177] (3) mission scenarios for the key information requirements generation process at mission decision points; and

[0178] (4) The branches driven by the decision point and the task scenarios for formulating subsequent plans.

[0179] As a preferred implementation, the prompt word template uses a generalized prompt word template as a basis, and an effective prompt contains one or more of the following five elements: role, instruction, context information, input data and output indication; a basic prompt word template is constructed based on historical data, and specific prompt word templates are designed based on the basic prompt word template for typical different intelligence task scenarios.

[0180] The prompt word template is the necessary glue to boost the large language model to output better. During the reasoning process, the large language model will retrieve one or more samples from the vector database as demonstrations based on the key information in the original question, and select the prompt word template suitable for the current scenario in the preset template library as the organizational structure of the input information, and combine the demonstration data and other information into a natural language prompt in a specific order with the selected specific task prompt word template. Finally, the natural language prompt is input into the fine-tuned large language model to form an accurate and standardized output that meets the commander's needs.

[0181] Combining existing prompt word technology, experience and practice, the present invention adopts a generalized prompt word template as the basis. A prompt can contain the following five elements, as shown in the following table. An effective prompt does not necessarily have to have all the elements, but is based on a specific analysis of the specific task. The following will be based on the basic prompt word template shown in Table 7 below, and will design specific prompt word templates for some typical intelligence task scenarios.

[0182] Table 7 Basic prompt word structure of this patent

[0183]

[0184]

[0185] As a preferred implementation, for the intelligence data retrieval, processing and application process task scenario, S12 includes:

[0186] In the field of modern military intelligence, the application of large language models is crucial to improving the efficiency of intelligence retrieval, processing, analysis and application. For military large language models fine-tuned with intelligence data, users can use some basic prompt word templates and structured input to allow the large model to understand user needs and produce answers that meet user expectations.

[0187] S121, determining a mapping relationship between each stage in combat intelligence preparation and the intelligence data retrieval, processing and application process task scenario; wherein the intelligence data processing includes intelligence processing and intelligence analysis;

[0188] In this embodiment:

[0189] 1. Intelligence data retrieval:

[0190] (1) Intelligence planning and guidance stage: In this stage, the large language model helps intelligence workers refine new search topics and understand the operational environment and potential enemy threats through its general knowledge ability and multi-round dialogue optimization. The model can quickly determine information related to possible enemy actions and improve the accuracy and relevance of intelligence collection. The key is to determine intelligence needs through efficient information retrieval, laying the foundation for subsequent intelligence collection and analysis.

[0191] (2) Intelligence collection phase: In this phase, the large language model helps analyze and filter the large amount of information collected. The model can process multilingual content, translate, summarize, and classify important information, and improve the availability and comprehensibility of intelligence. In addition, the model can also assist in identifying key time nodes, geographic locations, and enemy action types, making the collected intelligence more specific and targeted.

[0192] 2. Intelligence processing:

[0193] (1) Intelligence processing and utilization stage: In this stage, the analytical capabilities of large language models play a key role. The model can quickly analyze and integrate intelligence data obtained from various sensors and sources. The model helps identify important information gaps and proposes possible action plans and strategic recommendations through comparison and analysis. Through intelligent analysis, the model can also assist in identifying potential risks and opportunities and provide support for strategic decision-making.

[0194] (2) Intelligence analysis and production stage: In a joint combat environment, large language models can be used to generate intelligence products, such as combat environment analysis and enemy capability assessment reports. The model can generate customized intelligence products based on user needs and the commander's strategic intent, providing targeted intelligence support. At this stage, the model can also assist in completing document services, providing users with the required types of document information, including historical data, preliminary research, etc.

[0195] 3. Intelligence Analysis:

[0196] (1) Intelligence distribution and integration stage: In this stage, the large language model helps ensure the rapid and effective distribution and integration of intelligence. The model can assist in integrating intelligence into the decision support system. The model can also help various departments understand and use intelligence, ensuring the consistency and accuracy of intelligence.

[0197] (2) Intelligence assessment and feedback phase: This is an ongoing process, and the large language model can be used to evaluate the effectiveness and accuracy of intelligence at this stage. The model can assist in analyzing the effectiveness of intelligence collection and processing processes to ensure that the needs of commanders and joint forces are met.

[0198] 4. Intelligence Application:

[0199] In terms of intelligence applications, the application of big models is not limited to traditional document services and information product provision, but is based on content and knowledge output after understanding user intentions and related concepts. This includes but is not limited to document information provision, professional analysis, processing and handling of intelligence products.

[0200] The application of large language models in the field of modern military intelligence not only improves the efficiency and quality of intelligence processing, but also provides more comprehensive and in-depth support for military decision-making. In the above way, large language models can greatly improve the efficiency and quality of intelligence. In the stage of intelligence planning and guidance, the model lays a solid foundation for intelligence collection through precise retrieval and subject extraction. In the intelligence collection stage, the model's multilingual processing and information classification capabilities make the collected information more specific and targeted. In the stage of intelligence processing and utilization, the model's analysis and integration capabilities provide strong support for strategic decision-making. In the stage of intelligence analysis and production, the model not only accelerates the circulation and application of information, but also improves the customization and personalization of intelligence products.

[0201] S122, designing the prompt word template based on the mapping relationship;

[0202]

[0203]

[0204] As a preferred implementation, for the task decision point mining process task scenario, S12 includes:

[0205] S121', determining the application scenario of the task scenario of the task decision point mining process;

[0206] In modern military operations, the identification and analysis of mission decision points are crucial. For mission decision points, the large language model can comprehensively understand the impact of battlefield environment, such as weather factors, on combat operations, recommend locations suitable for various types of equipment (including drones, ships, etc.) to perform combat missions, develop one's own action plans, and assist commanders in making decisions.

[0207] Mission decision points are divided into expected mission decision points and potential mission decision points. For expected mission decision points, the traditional method uses key event analysis to clarify the expected decision points, because no matter what type of military action is taken and what kind of combat environment is in, and no matter how many variables may be encountered during the implementation of the action, some key events are inevitable in the early stage of the action, and the tasks that must be completed are also relatively certain. Decision points are associated with key events, and expected decision points are set for each key event. These key events and expected decision points that are determined to occur are analyzed and deduced in detail, so that when the action implementation progresses to the decision point, the commander and his staff team can ensure the synchronization and coordination of the action forces and resources, ensure the expected action effect, and guide the situation to the development track of their own actions. For expected mission decision points, the traditional method discovers potential decision points through action plan war game deduction. War game deduction of action plans can enable commanders at all levels and their staff teams to better understand their own action plans and possible threats in the plans, so as to scientifically judge the advantages and disadvantages of each action plan, and lay the foundation for subsequent action plan comparison and action plan approval.

[0208] As technology advances, large language models (LLMs) provide a novel approach to assist this process, especially in mining accurate decision point information using cue word engineering. This patented design combines traditional methods with LLM's cue word technology to create a more efficient and in-depth decision point mining process.

[0209] S122', designing the prompt word template based on the application scenario of the task scenario in the task decision point mining process and the prompt word technology of LLM;

[0210] First, using LLM’s powerful data processing capabilities, we can quickly analyze a large amount of historical military operations and strategic decision-making data. This includes not only public historical records, but also internal reports, communication records, and intelligence data. By designing targeted prompt words for the model, we can guide the model to identify and analyze cases that have been considered critical events in the past, thereby helping decision makers identify possible expected decision points.

[0211] In addition, in terms of mining potential decision points, LLM's prompt word technology can simulate different battlefield scenarios and action plans. Compared with traditional war games, LLM can generate and analyze various possible scenarios more quickly, thereby revealing potential decision points that may not be noticed in traditional methods. For example, by setting specific battlefield conditions or enemy behaviors as prompt words, LLM can generate multiple scenarios related to these conditions, helping decision makers explore and identify key decision points that may arise in different situations. In addition, using the capabilities of LLM, we can also analyze the decision-making needs of specific situations in more depth. For example, combined with geographic information system (GIS) data and real-time intelligence, LLM can provide detailed analysis of specific places and times, helping decision makers to locate decision points more accurately.

[0212] In practical applications, LLM's prompt word technology can be combined with traditional war game simulations. Traditional simulations provide judgments and tactical considerations based on experience, while LLM provides in-depth analysis based on large amounts of data. Through this combination, we can not only improve the efficiency of decision point mining, but also improve its accuracy and comprehensiveness, so that the model can not only process large amounts of historical data and simulate complex battlefield scenarios, but also provide in-depth analysis in real time, thereby helping decision makers make more accurate and effective decisions in complex and dynamic military environments.

[0213] The example of the prompt word template designed in this embodiment is as follows:

[0214]

[0215] As a preferred implementation, for the task scenario of the key information requirement generation process for the task decision point, S12 includes:

[0216] S121”, determine the application scenario of the task scenario of the key information requirement generation process for the task decision point;

[0217] In military operations, commanders' decisions are often faced with time and space urgency, which requires commanders to make critical decisions at specific times and locations, and these decisions rely on accurate and timely intelligence support.

[0218] S122”, designing the prompt word template based on the application scenario of the task scenario of the key information requirement generation process for the task decision point and natural language processing and deep learning technology;

[0219] In this embodiment, the application of large language model prompt word engineering in the process of generating key information requirements for task decision points includes:

[0220] The commander's information needs are mainly divided into two categories: priority intelligence needs and friendly force information needs. Priority intelligence needs focus on the enemy or potential threats, involving the formulation of intelligence collection plans, the clarification of intelligence forces and task priorities. Friendly force information needs involve the status, capabilities and action plans of friendly forces. The practice of the US military shows that these information needs are crucial to ensuring the effectiveness of battlefield decision-making. The intelligence department plays a dual role in combat: one is to provide intelligence support for combat planning; the other is to plan intelligence reconnaissance operations according to combat needs. In careful action planning, the intelligence department prepares for possible emergencies through a complete intelligence planning and support line. In crisis action planning, due to time constraints, intelligence support and planning content need to be tailored to fit the combat plan.

[0221] In this context, the empowerment of large language model prompt word engineering becomes crucial. This technology can help intelligence analysts quickly generate key information requirements related to specific decision points. Through natural language processing and deep learning technology, the model can: Based on historical intelligence data and past actions, the model can predict future information needs to guide the priority of intelligence collection; identify potential patterns and trends in enemy actions to provide strategic insights for commanders; as the battlefield situation changes, the model can update information needs in real time to ensure that commanders obtain the latest intelligence; evaluate the potential risks and benefits of different decision-making options to assist commanders in making more scientific decisions.

[0222] The example of the prompt word template designed in this embodiment is as follows:

[0223]

[0224]

[0225] As a preferred implementation, for the branch driven by the decision point and the subsequent plan formulation task scenario, the S12 includes:

[0226] S121”′, determining the application scenario of the branch driven by the decision point and the formulation task scenario of the subsequent plan;

[0227] In the modern military combat environment, decision-making and intelligence processing are two important links that complement each other. Especially in the complex and ever-changing battlefield environment, the formulation of strategic decisions often needs to be based on a large amount of accurate and timely intelligence information.

[0228] S122'', designing the prompt word template based on the branches driven by the decision point and the application scenario of the subsequent plan formulation task scenario.

[0229] In this embodiment, the deduction and revision of the other party's action plan and the intelligence demand model for strategic decision-making provide two important cases to help the present invention understand how to use the large language model prompt word engineering empowerment to carry out decision-point-driven branching and subsequent plan formulation under high pressure and high uncertainty conditions.

[0230] First, consider the process of deducing and revising the opponent's action plan. In actual operation, after the first deduction of the action plan, some deficiencies and potential decision points will be exposed. At this time, the staff team needs to adjust and optimize the plan to improve its pertinence and flexibility. In this process, the big language model can play a key role. By analyzing historical data, battlefield environment and possible hostile actions, the model can help decision makers identify potential weaknesses and opportunities. For example, if the model predicts that the enemy may change the direction of attack or new threats may emerge, the staff team can formulate branch plans and follow-up plans accordingly.

[0231] In practical applications, the branching and subsequent planning driven by the decision point usually include the following steps:

[0232] Intelligence collection and analysis: Use large language models to analyze collected data and identify key decision points, such as enemy potential

[0233] A strategic weakness in action or one's own side.

[0234] ·Scheme deduction and revision: Based on the analysis results provided by the model, conduct preliminary deduction of the scheme and identify the areas that need to be revised.

[0235] part.

[0236] ·Branch plan formulation: formulate branch plans for the identified key decision points to deal with possible battlefield changes or sudden

[0237] Send the situation.

[0238] Follow-up planning: Consider the possibility of different battle outcomes, such as victory, defeat, or stalemate, and develop a plan for each scenario.

[0239] Appropriate follow-up action plan.

[0240] Real-time adjustment and optimization: During the actual operation, the branches are constantly revised and optimized according to the real-time changes in the battlefield situation.

[0241] and subsequent plans.

[0242] Decision execution and feedback: Execute decisions and continuously optimize models and decision-making processes based on feedback from action results.

[0243] In this way, the big language model not only provides support in the intelligence analysis stage, but also plays a key role in the entire decision-making process. Through the structured prompt word template and partial thinking chain thinking, the model predicts and simulates different battlefield situations in steps, helping decision makers to better understand and predict possible changes, so as to make more flexible and adaptable decisions. This decision-making process not only improves the success rate of actions, but also enables rapid adjustment of strategies to cope with the ever-changing battlefield environment when responding to emergencies.

[0244] The example of the prompt word template designed in this embodiment is as follows:

[0245]

[0246]

[0247] S2, understanding the intent of key intelligence requirements based on the key intelligence requirement intention understanding big model; wherein the intelligence-based decision-making includes associating commanders' key information requirements generation, mining mission decision points, and specifying traction branches and subsequent plans.

[0248] Embodiment 2

[0249] like Figure 6 As shown, this embodiment provides a key intelligence demand intention understanding system based on a large model, which is used to implement the method described in the first aspect, including:

[0250] Model building module 101, used to build a key intelligence demand intention understanding big model; wherein the key intelligence demand intention understanding big model includes interactive instruction understanding algorithm, combat plan understanding algorithm, contextual prompt learning technology, heterogeneous intelligence association fusion algorithm and multimodal intelligence understanding algorithm;

[0251] The key intelligence requirement intention understanding module 102 is used to understand the key intelligence requirement intention based on the key intelligence requirement intention understanding large model; wherein the intelligence-based decision-making includes associating the commander's key information requirements, mining task decision points, and specifying traction branches and subsequent plans.

[0252] The core invention of this embodiment is:

[0253] Method for constructing an intelligence vector database and an intelligence command fine-tuning dataset. This method designs a real-time updated intelligence vector database that collects historical intelligence mission data, including inquiries, reports, analysis results, etc., and retrieves information quickly and efficiently through advanced data processing technologies (such as vectorization and indexing). In addition, an expert-generated intelligence command fine-tuning dataset is designed and constructed for multimodal command fine-tuning of basic military large-scale language models. With the support of this solution, the timeliness and relevance of intelligence data are ensured, the understanding and response capabilities of large models for specific military tasks are improved, and the retrieval, processing and application processes of intelligence data are optimized. An intelligence vector database and an intelligence command fine-tuning dataset are constructed to ensure real-time information updates and quality control. In the task reasoning stage, the model dynamically adjusts the focus and direction of analysis by quickly retrieving samples in the intelligence vector database as context resources.

[0254] A chatglm-based large model fine-tuning method. Use the intelligence command fine-tuning dataset to fine-tune the basic military large language model for specific intelligence tasks to enhance the model's understanding and response accuracy for specific types of military tasks. Ultimately, improve the model's performance in different mission scenarios, enhance the model's ability to adjust real-time data of generated content, and promote mission decision point mining and key intelligence demand generation. Based on ChatGLM technology, perform efficient parameter fine-tuning on the military large language model to enhance the model's understanding and response capabilities for specific military tasks.

[0255] Intelligence mission planning prompt template driven by mission decision points. Through the flexible construction of prompt word templates, customized information filtering and sorting options are provided, and examples and other information are organically combined to form prompt inputs for large language models. The model output is adapted to the preferences and needs of different commanders, and the information display and analysis methods can be automatically adjusted according to the current mission and environment, improving the efficiency of intelligence writing and human-computer interaction. The preset prompt word template library realizes the customization and systematization of intelligence mission planning, ensures the accuracy and efficiency of input, automatically generates high-quality intelligence that meets the current mission requirements, and provides efficient information filtering and sorting options to enhance the close integration of intelligence and decision-making.

[0256] The present invention also provides a memory storing a plurality of instructions, wherein the instructions are used to implement the method as in the first embodiment.

[0257] like Figure 7 As shown, the present invention also provides an electronic device, including a processor 301 and a memory 302 connected to the processor 301, the memory 302 stores a plurality of instructions, and the instructions can be loaded and executed by the processor so that the processor can execute the method as in the first embodiment.

[0258] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for understanding key intelligence requirements based on a large model, characterized in that: include: S1, building a big model for understanding key intelligence requirements and intentions; The key intelligence demand intention understanding big model includes interactive instruction understanding algorithm, combat plan understanding algorithm, contextual prompt learning technology, heterogeneous intelligence association fusion algorithm and multimodal intelligence understanding algorithm; the key intelligence demand intention understanding big model is based on the automated intelligence analysis and understanding of the joint big model efficient parameter fine-tuning, retrieval enhancement generation and prompt word technology; S2, understanding the intent of key intelligence requirements based on the key intelligence requirement intention understanding big model; wherein the intelligence-based decision-making includes associating commanders' key information requirements generation, mining mission decision points, and specifying traction branches and subsequent plans.

2. According to the method of understanding key intelligence requirements based on a large model according to claim 1, it is characterized in that: The S1 includes: S11, fine-tuning the basic military large language model enabled by chatglm through initial intelligence data to obtain a fine-tuned model; S12, based on the fine-tuning model and based on the prompt word engineering technology, construct a prompt word template for a specific field, so that the fine-tuning model can meet the needs of different intelligence tasks.

3. According to the method of understanding key intelligence requirements based on a large model according to claim 2, it is characterized in that: The S11 includes: S111, construct three types of data sets, namely, intelligence instruction fine-tuning data set, intelligence vector database and preset prompt word template library; wherein: The intelligence instruction fine-tuning dataset is used for fine-tuning the military large language model to enhance the model's ability to understand and respond to specific types of military tasks. It is a high-quality fine-tuning dataset designed and generated based on specific military intelligence tasks, and contains task-related questions and answers, as well as specific scenarios, terms, protocols and other elements. The intelligence vector database integrates historical intelligence mission data, including past inquiries, reports, analysis results, etc., and integrates new data generated by experts into the database to ensure the timeliness and relevance of information; The preset prompt word template library stores multiple templates designed in advance for different types of intelligence tasks to ensure the accuracy and efficiency of input; S112, for a new original question, first perform keyword extraction to identify the core elements and themes of the question, and based on these keywords, select the most appropriate template from a preset prompt word template library; then, based on the intelligence instruction fine-tuning data set, use efficient parameter fine-tuning technology to fine-tune the basic military large language model to form a military large language model adapted to the intelligence field; S113, fine-tuning large military models adapted to the intelligence field based on efficient parameter fine-tuning methods.

4. According to claim 3, a method for understanding key intelligence requirements based on a large model is characterized in that: The S113 includes: (1) Collect and filter data to obtain intelligence data from different data sources and data modalities; including: Collecting raw intelligence data: Collecting raw data from one or more data sources including open source intelligence, human intelligence, signals intelligence, geospatial intelligence, and technical intelligence; Screening of raw intelligence data: performing data cleaning, data verification, data classification and labeling, data analysis, data fusion, and confidentiality and ethical processing on the raw intelligence data, and collecting and screening key information from the multi-source and multi-modal intelligence data as the data basis for the training of the large model and decision support; (2) constructing multiple data models based on the integration of the filtered intelligence data and the characteristics and requirements of each intelligence; wherein the multiple data models correspond to each intelligence category one by one and are used to manage and analyze complex intelligence data; the data models include: Entity-relationship model (ERM) and text analysis model for open source intelligence; Graph database models and time-driven models for human intelligence; Time series models and network traffic analysis models for signals intelligence; Geographic Information System (GIS) model and multidimensional data model (OLAP) for geospatial intelligence; Database models and association rule learning models for technical intelligence; (3) constructing different types of data sets based on the constructed data model; wherein the data sets include text understanding data sets, intelligence-related data sets, decision point mining data sets, and auxiliary task planning data sets; The text understanding datasets include: interactive instruction understanding datasets and multimodal intelligence understanding datasets; The intelligence-related data sets include: intelligence report data sets, operational intelligence requirements understanding data sets, and intelligence mission planning and decomposition data sets; The decision point mining data set includes: a decision point case data set and a scenario simulation data set; The auxiliary mission planning data set includes: command and control communication data set and tactics and strategy guidance data set; (4) constructing an intelligence vector data set and an expected intelligence instruction fine-tuning data set based on the data set, wherein the intelligence vector data set is updated in real time and is stored in real time whenever there is the latest real intelligence data; the expected intelligence instruction fine-tuning data set is stored in a data architecture that complies with the instruction fine-tuning specification after being screened, checked and supplemented by military experts, and is used to fine-tune the basic military large language model. The expected intelligence instruction fine-tuning data set is updated periodically according to the fine-tuning requirements; (5) Fine-tune the military large model adapted to the intelligence field based on multi-modal instruction fine-tuning technology and efficient parameter fine-tuning technology; The multimodal instruction fine-tuning technology is used to generate an intelligence instruction fine-tuning dataset based on prompt templates and experts through a large language, wherein the intelligence instruction fine-tuning dataset includes a seed instruction set and a corresponding input and output instance set to generate a large amount of new instruction data and corresponding input and output instances, and then use the generated instruction dataset to fine-tune the instructions of downstream tasks on the large model; The efficient parameter fine-tuning technology includes setting a model and a word segmenter and selecting a suitable parameter quantization model according to the current fine-tuning data volume.

5. According to the method of understanding key intelligence requirements based on a large model according to claim 4, it is characterized in that: The step S12, based on the fine-tuning model and based on the prompt word engineering technology, constructs a prompt word template for a specific field so that the fine-tuning model meets the needs of different intelligence tasks, including: constructing the prompt word template for different intelligence task scenarios; the different intelligence task scenarios include: (1) Mission scenarios for intelligence data retrieval, processing, and application; (2) Task decision point mining process task scenario; (3) mission scenarios for the key information requirements generation process at mission decision points; and (4) The branches driven by the decision point and the task scenarios for formulating subsequent plans.

6. According to the method of understanding key intelligence requirements based on a large model according to claim 5, it is characterized in that: The prompt word template uses a generalized prompt word template as a basis. An effective prompt contains one or more of the following five elements: role, instruction, context information, input data and output indication. A basic prompt word template is constructed based on historical data, and specific prompt word templates are designed based on the basic prompt word template for typical different intelligence task scenarios.

7. According to claim 6, a method for understanding key intelligence requirements based on a large model is characterized in that: For the intelligence data retrieval, processing and application process task scenario, S12 includes: S121, determine the mapping relationship between each stage in combat intelligence preparation and the task scenario of the intelligence data retrieval, processing and application process; wherein the processing of the intelligence data includes intelligence processing and intelligence analysis; for intelligence data retrieval, it corresponds to the intelligence plan formulation and guidance stage and the intelligence collection stage; for intelligence processing, it corresponds to the intelligence processing and utilization stage and the intelligence analysis and production stage; for intelligence analysis, it corresponds to the intelligence distribution and fusion stage and the intelligence evaluation and feedback stage; for intelligence application, it corresponds to the traditional document service and intelligence product provision stage and the content and knowledge output stage based on understanding user intentions and related concepts, including document information provision, professional analysis, processing and handling of intelligence products; S122, designing the prompt word template based on the mapping relationship; For the task decision point mining process task scenario, S12 includes: S121', determining the application scenario of the task scenario of the task decision point mining process; S122', designing the prompt word template based on the application scenario of the task scenario in the task decision point mining process and the prompt word technology of LLM; For the task scenario of the key information requirement generation process for the task decision point, S12 includes: S121”, determine the application scenario of the task scenario of the key information requirement generation process for the task decision point; S122”, designing the prompt word template based on the application scenario of the task scenario of the key information requirement generation process for the task decision point and natural language processing and deep learning technology; For the branch driven by the decision point and the subsequent plan formulation task scenario, S12 includes: S121”′, determining the application scenario of the branch driven by the decision point and the formulation task scenario of the subsequent plan; S122'', designing the prompt word template based on the branches driven by the decision point and the application scenario of the subsequent plan formulation task scenario.

8. A key intelligence requirement intention understanding system based on a large model, used to implement any of the methods of claims 1-7, comprising: A model building module (101) is used to build a big model for understanding key intelligence requirements and intentions; The key intelligence demand intention understanding model includes interactive instruction understanding algorithm, combat plan understanding algorithm, contextual prompt learning technology, heterogeneous intelligence association fusion algorithm and multimodal intelligence understanding algorithm; A key intelligence requirement intention understanding module (102) is used to understand the intention of key intelligence requirements based on the key intelligence requirement intention understanding large model; wherein the module is used for intelligence-based decision-making, and the intelligence-based decision-making includes associating commanders with key information requirements, mining mission decision points, and specifying traction branches and subsequent plans.

9. An electronic device, characterized in that: The method comprises a processor and a memory, wherein the memory stores a plurality of instructions, and the processor is used to read the instructions and execute the method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a plurality of instructions, and the plurality of instructions can be read by a processor to execute the method according to any one of claims 1 to 7.

Images