Method and device for generating target object analysis report

Abstract

The present disclosure provides a method and device for generating a target object analysis report. The method includes: obtaining a training corpus set, wherein the training corpus set contains multiple training corpora about different objects; using multiple training corpora to incrementally train a large language model that has been pre-trained, so that the large language model learns relevant knowledge about the objects corresponding to each training corpus; generating multiple analysis instructions, and using multiple training corpora to train the incrementally trained large language model according to each analysis instruction, respectively, to obtain a task model corresponding to each analysis instruction; when it is necessary to analyze the target object, retrieve the text corpus related to the target object; and collaboratively process the text corpus through the task models corresponding to each analysis instruction to form a comprehensive analysis report on the target object. The above technical means can solve the problem of low efficiency and inaccuracy in commodity information analysis in the existing technology, thereby improving the efficiency and accuracy of commodity information analysis.

Description

Technical Field

[0001] The present disclosure relates to the field of text processing technology, and in particular to a method and device for generating a target object analysis report. Background Art

[0002] In the marketing and retail industries, product analysis capabilities are crucial to business success. Companies often need to identify hidden business opportunities from vast amounts of market, competitor, and consumer data, adjusting marketing strategies and optimizing product portfolios accordingly. However, this manual process requires significant time and effort to collect and analyze data, resulting in inefficiency and prone to bias. Leveraging artificial intelligence to automatically extract valuable product insights from diverse data sources has long been a major challenge in this field. Summary of the Invention

[0003] In view of this, embodiments of the present disclosure provide a method, apparatus, electronic device, and computer-readable storage medium for generating a target object analysis report to solve the problem of low efficiency and inaccuracy in commodity information analysis in the prior art.

[0004] In a first aspect of an embodiment of the present disclosure, a method for generating a target object analysis report is provided, comprising: obtaining a training corpus set, wherein the training corpus set includes multiple training corpora about different objects; performing incremental training on a pre-trained large language model using the multiple training corpora, so that the large language model learns relevant knowledge about the objects corresponding to each training corpus; generating multiple analysis instructions, and training the incrementally trained large language model using the multiple training corpora according to each analysis instruction, to obtain a task model corresponding to each analysis instruction; when it is necessary to analyze the target object, retrieving text corpora related to the target object; and collaboratively processing the text corpora through the task models corresponding to each analysis instruction to form a comprehensive analysis report on the target object.

[0005] According to a second aspect of an embodiment of the present disclosure, there is provided an apparatus for generating an analysis report on a target object, comprising: an acquisition module configured to acquire a training corpus, wherein the training corpus includes a plurality of training corpora regarding different objects; an incremental training module configured to perform incremental training on a pre-trained large language model using the plurality of training corpora, so that the large language model learns relevant knowledge about the objects corresponding to the respective training corpora; an analysis training module configured to generate a plurality of analysis instructions, and train the incrementally trained large language model using the plurality of training corpora according to each analysis instruction, to obtain a task model corresponding to each analysis instruction; a retrieval module configured to retrieve text corpora related to the target object when analysis of the target object is required; and a collaborative analysis module configured to collaboratively process the text corpora through the task models corresponding to the respective analysis instructions, so as to form a comprehensive analysis report on the target object.

[0006] According to a third aspect of an embodiment of the present disclosure, an electronic device is provided, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the above method when executing the computer program.

[0007] According to a fourth aspect of the embodiments of the present disclosure, a computer-readable storage medium is provided, which stores a computer program. When the computer program is executed by a processor, the steps of the above method are implemented.

[0008] Compared with the prior art, the embodiments of the present disclosure have the following beneficial effects: obtaining a training corpus, wherein the training corpus contains multiple training corpora about different objects; using multiple training corpora to incrementally train a large language model that has been pre-trained, so that the large language model learns relevant knowledge about the objects corresponding to each training corpus; generating multiple analysis instructions, and using multiple training corpora to train the incrementally trained large language model according to each analysis instruction, to obtain a task model corresponding to each analysis instruction; when it is necessary to analyze the target object, retrieve the text corpus related to the target object; and collaboratively process the text corpus through the task models corresponding to each analysis instruction to form a comprehensive analysis report on the target object. The above technical means can solve the problem of low efficiency and inaccuracy in commodity information analysis in the prior art, thereby improving the efficiency and accuracy of commodity information analysis. BRIEF DESCRIPTION OF THE DRAWINGS

[0009] In order to more clearly illustrate the technical solutions in the embodiments of the present disclosure, the following briefly introduces the drawings required for use in the embodiments or descriptions of the prior art. Obviously, the drawings in the following description are only some embodiments of the present disclosure. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0010] Figure 1 This is a flowchart of a method for generating a target object analysis report provided by an embodiment of the present disclosure;

[0011] Figure 2 This is a flowchart of another method for generating a target object analysis report provided by an embodiment of the present disclosure;

[0012] Figure 3 Schematic diagram of a device for generating a target object analysis report according to an embodiment of the present disclosure;

[0013] Figure 4 This is a schematic diagram of the structure of an electronic device provided by an embodiment of the present disclosure. DETAILED DESCRIPTION

[0014] In the following description, specific details such as particular system structures and techniques are provided for purposes of illustration, not limitation, to facilitate a thorough understanding of the embodiments of the present disclosure. However, it will be apparent to those skilled in the art that other embodiments of the present disclosure may be implemented without these specific details. In other cases, detailed descriptions of well-known systems, devices, circuits, and methods are omitted to avoid obscuring the present disclosure with unnecessary detail.

[0015] A method and apparatus for generating a target object analysis report according to an embodiment of the present disclosure will be described in detail below with reference to the accompanying drawings.

[0016] Figure 1 The present invention provides a flowchart of a method for generating a target object analysis report. Figure 1 The method for generating the target object analysis report can be executed by a computer or a server, or by software on the computer or the server. Figure 1 As shown, the method for generating a target object analysis report includes:

[0017] S101, obtaining a training corpus, wherein the training corpus includes a plurality of training corpora on different objects;

[0018] S102, incrementally training the pre-trained large language model using multiple training corpora, so that the large language model learns relevant knowledge about the object corresponding to each training corpus;

[0019] S103, generating multiple analysis instructions, and using multiple training corpora to train the incrementally trained large language model according to each analysis instruction, to obtain a task model corresponding to each analysis instruction;

[0020] S104, when the target object needs to be analyzed, retrieve text corpus related to the target object;

[0021] S105 , collaboratively processing the text corpus through the task models corresponding to the various analysis instructions to form a comprehensive analysis report on the target object.

[0022] The training corpus contains multiple training data items, each of which is a text corpus about an object, such as a product. The training corpus is collected from various product-related text corpora on the internet, including news reports, industry reports, technical white papers, and social media. This raw data is obtained through preprocessing such as data cleaning and segmentation. The training data includes product attribute information, user reviews, and price change records. Product attribute information includes category, function, name, and so on.

[0023] Large language models are pre-trained based on common text processing tasks using standard corpora. This pre-trained large language model can predict subsequent text based on a single or multiple consecutive texts. Llama-2 is a large language model based on the Transformer. It achieves excellent performance through a series of optimizations, such as pre-training with more data, using longer context lengths, and adopting an architecture optimized for fast inference.

[0024] Analyzing the target object requires multifaceted analysis. Therefore, during the training phase, multiple analysis instructions are generated. Based on each analysis instruction, the incrementally trained large language model is trained using multiple training corpora, resulting in a corresponding task model. During the inference phase, the task model corresponding to each analysis instruction performs a comprehensive analysis of the target object's text corpus, ultimately generating a comprehensive analysis report on the target object.

[0025] According to the technical solution provided in the embodiment of the present application, a training corpus is obtained, wherein the training corpus contains multiple training corpora about different objects; the large language model that has been pre-trained is incrementally trained using the multiple training corpora, so that the large language model learns relevant knowledge about the objects corresponding to each training corpus; multiple analysis instructions are generated, and the large language model after incremental training is trained using the multiple training corpora according to each analysis instruction, so as to obtain the task model corresponding to each analysis instruction; when it is necessary to analyze the target object, the text corpus related to the target object is retrieved; the text corpus is collaboratively processed by the task models corresponding to each analysis instruction to form a comprehensive analysis report on the target object. The above-mentioned technical means can solve the problem of low efficiency and inaccuracy of commodity information analysis in the prior art, thereby improving the efficiency and accuracy of commodity information analysis.

[0026] Furthermore, the pre-trained large language model is incrementally trained using multiple training corpora, so that the large language model learns relevant knowledge about the objects corresponding to each training corpus, including: the incremental training is contextual training, and the large language model predicts the subsequent words of a word or multiple consecutive words based on the word or multiple consecutive words in each training corpus; the large language model learns relevant knowledge about the objects corresponding to each training corpus through incremental training.

[0027] The methods of incremental training and pre-training are similar, the difference lies in the different training corpus used.

[0028] Incremental training gives the large language model a certain degree of product knowledge understanding ability. Product knowledge understanding ability includes the following aspects: (1) Product knowledge understanding ability refers to an in-depth understanding and mastery of various product-related knowledge, including but not limited to: (2) Product classification system: being able to master the product classification system of various industry standards and understand what specific product types are included in each category. (3) Product attributes and specifications: having a thorough understanding of the key attributes and specifications of various products, such as the hardware configuration of electronic products, the size and material of clothing, etc. (4) Product functions and uses: accurately grasping the design functions, typical application scenarios and usage methods of products. (5) Product brands and origins: being familiar with the main brands, production areas and brand characteristics of different products. (6) Product price levels: having a reasonable judgment on the normal price range of different categories of products. (7) Product knowledge association: being able to link product knowledge with related fields (such as technology, materials, aesthetics, etc.). (8) Product development trends: having insight into the future development direction of product design and market.

[0029] Furthermore, according to each analysis instruction, a plurality of training corpora are used to train the incrementally trained large language model to obtain a task model corresponding to each analysis instruction, including: for each analysis instruction: constructing the labeled data under the analysis instruction for each training corpus; using the analysis instruction as a prompt word, inputting the prompt word and each training corpus into the incrementally trained large language model, and outputting the prediction result corresponding to each training corpus; calculating the loss between the prediction result corresponding to each training corpus and the labeled data, optimizing the model parameters of the large language model according to the loss corresponding to each training corpus, and obtaining the task model corresponding to the analysis instruction.

[0030] The cross entropy loss function can be used to calculate the loss between the predicted results and the labeled data corresponding to each training corpus.

[0031] Furthermore, according to each analysis instruction, the incrementally trained large language model is trained using multiple training corpora to obtain a task model corresponding to each analysis instruction, including: each analysis instruction, including: object attribute information extraction instruction, object user comment information mining instruction, object price trend analysis instruction and object development prospect analysis instruction; according to the object attribute information extraction instruction, the incrementally trained large language model is trained using multiple training corpora to obtain a task model corresponding to the object attribute information extraction instruction; according to the object user comment information mining instruction, the incrementally trained large language model is trained using multiple training corpora to obtain a task model corresponding to the object user comment information mining instruction; according to the object price trend analysis instruction, the incrementally trained large language model is trained using multiple training corpora to obtain a task model corresponding to the object price trend analysis instruction; according to the object development prospect analysis instruction, the incrementally trained large language model is trained using multiple training corpora to obtain a task model corresponding to the object development prospect analysis instruction.

[0032] The labeled data of the training corpus under different analysis instructions are different. The labeled data of the training corpus under the object attribute information extraction instruction is the object attribute information labeled in advance. The labeled data of the training corpus under the object user comment information mining instruction is the user comment information labeled in advance. The labeled data of the training corpus under the object price trend analysis instruction is the object price trend labeled in advance. The labeled data of the training corpus under the object development prospect analysis instruction is the object development prospect information labeled in advance. For example, under the object attribute information extraction instruction, the object attribute information extraction instruction is used as a prompt word, and the object attribute information extraction instruction and each training corpus are respectively input into the large language model after incremental training, and the prediction results corresponding to each training corpus are output; the loss between the prediction results corresponding to each training corpus and the labeled data is calculated, and the model parameters of the large language model are optimized according to the loss corresponding to each training corpus. The optimized ceiling large language model is used as the task model corresponding to the object attribute information extraction instruction.

[0033] The task model corresponding to the object attribute information extraction instruction is used to extract the attribute information of the product; the task model corresponding to the object user review information mining instruction is used to extract and summarize the user review information of the product; the task model corresponding to the object price trend analysis instruction is used to analyze the price trend of the product; the task model corresponding to the object development prospect analysis instruction is used to analyze the development prospect of the product.

[0034] Furthermore, the text corpus is collaboratively processed by the task models corresponding to each analysis instruction to form a comprehensive analysis report on the target object, including: processing the text corpus in parallel by the task models corresponding to each analysis instruction to obtain analysis results corresponding to each analysis instruction, wherein, in the process of processing the text corpus, each task model exchanges analysis results with each other at a preset frequency, and uses the analysis results of other task models as reference information to update its own analysis results; and summarizing the analysis results corresponding to each analysis instruction to form a comprehensive analysis report on the target object.

[0035] As each task model processes the text corpus, it regularly exchanges partial results and findings, learning from and inspiring each other. Finally, a comprehensive analysis report on the target object is generated based on the analysis results corresponding to each analysis instruction.

[0036] Furthermore, the large language model includes a vocabulary network, which is used to convert the input of the large language model into a specific sequence. Other networks in the large language model process the sequence output by the vocabulary network to obtain the output of the large language model; within the vocabulary network: the input of the large language model is divided into a sequence consisting of single words; the frequency of occurrence of each word is counted, the words that appear most frequently are merged into a new symbol, and the sequence of the words that appear most frequently is updated using the new symbol; through continuous iteration until the number of mergers reaches a preset number, the merged sequence is used as the sequence output by the vocabulary network, wherein each iteration merges the symbols that appear most frequently and updates the sequence of the symbols that appear most frequently.

[0037] For example, if we count all the words that appear most frequently, and "l" and "o" appear the most frequently, we can merge "l" and "o" into a new symbol "lo", and then replace all "l" and "o" in the original sequence with "lo". Merging the most frequently appearing symbols is similar to merging the most frequently appearing words.

[0038] Other networks, including: encoding network, decoding network, attention network and feedforward neural network;

[0039] The encoding network is used to encode the sequence to extract the semantic information of the sequence; the decoding network is used to decode the semantic information to obtain decoded information; the attention network is used to learn the long-distance dependency of the decoded information through a self-attention mechanism; and the feedforward neural network is used to perform nonlinear transformation on the input sequence to enhance the expressive power of the model.

[0040] Figure 2 FIG. 1 is a flow chart of another method for generating a target object analysis report provided by an embodiment of the present disclosure. Figure 2 As shown, the method includes:

[0041] Various analysis instructions, including: object attribute information extraction instructions, object user comment information mining instructions, object price trend analysis instructions and object development prospect analysis instructions;

[0042] S201, processing text corpus through the task model corresponding to the object attribute information extraction instruction to obtain the analysis result corresponding to the object attribute information extraction instruction;

[0043] S202, obtaining analysis results corresponding to the user review information mining instructions of the object through a task model corresponding to the user review information mining instructions of the object;

[0044] S203, obtaining an analysis result corresponding to the object price trend analysis instruction through a task model corresponding to the object price trend analysis instruction;

[0045] S204, obtaining an analysis result corresponding to the object development prospect analysis instruction through a task model corresponding to the object development prospect analysis instruction;

[0046] S205: Summarize the analysis results corresponding to the various analysis instructions to form a comprehensive analysis report on the target object.

[0047] The task model for the object attribute information extraction instruction is used to extract the attribute information of the target object; the task model for the object user review information mining instruction is used to extract and summarize the user reviews of the target object; the task model for the object price trend analysis instruction is used to analyze the price trend of the target object; and the task model for the object development prospect analysis instruction is used to analyze the development prospect of the target object. The results of each analysis are summarized to form a comprehensive analysis report on the target object.

[0048] All of the above optional technical solutions can be combined in any way to form optional embodiments of the present application, and will not be described in detail here.

[0049] The following are embodiments of the apparatus disclosed herein, which can be used to implement the method embodiments disclosed herein. For details not disclosed in the apparatus embodiments disclosed herein, please refer to the method embodiments disclosed herein.

[0050] Figure 3 FIG. 1 is a schematic diagram of a device for generating a target object analysis report according to an embodiment of the present disclosure. Figure 3 As shown, the device for generating a target object analysis report includes:

[0051] An acquisition module 301 is configured to acquire a training corpus, wherein the training corpus includes a plurality of training corpora related to different objects;

[0052] Incremental training module 302 is configured to incrementally train the pre-trained large language model using multiple training corpora, so that the large language model learns relevant knowledge about the object corresponding to each training corpus;

[0053] The analysis and training module 303 is configured to generate a plurality of analysis instructions, and train the incrementally trained large language model using a plurality of training corpora according to each analysis instruction, to obtain a task model corresponding to each analysis instruction;

[0054] The retrieval module 304 is configured to retrieve text corpus related to the target object when the target object needs to be analyzed;

[0055] The collaborative analysis module 305 is configured to collaboratively process the text corpus through the task models corresponding to the various analysis instructions to form a comprehensive analysis report on the target object.

[0056] According to the technical solution provided in the embodiment of the present application, a training corpus is obtained, wherein the training corpus contains multiple training corpora about different objects; the large language model that has been pre-trained is incrementally trained using the multiple training corpora, so that the large language model learns relevant knowledge about the objects corresponding to each training corpus; multiple analysis instructions are generated, and the large language model after incremental training is trained using the multiple training corpora according to each analysis instruction, so as to obtain the task model corresponding to each analysis instruction; when it is necessary to analyze the target object, the text corpus related to the target object is retrieved; the text corpus is collaboratively processed by the task models corresponding to each analysis instruction to form a comprehensive analysis report on the target object. The above-mentioned technical means can solve the problem of low efficiency and inaccuracy of commodity information analysis in the prior art, thereby improving the efficiency and accuracy of commodity information analysis.

[0057] In some embodiments, the incremental training module 302 is further configured to perform incremental training as contextual training, where the large language model predicts the subsequent words of a word or multiple consecutive words based on a word or multiple consecutive words in each training corpus; the large language model learns relevant knowledge about the objects corresponding to each training corpus through incremental training.

[0058] In some embodiments, the analysis and training module 303 is further configured to, for each analysis instruction: construct the labeled data under the analysis instruction for each training corpus; use the analysis instruction as a prompt word, input the prompt word and each training corpus into the large language model after incremental training, and output the prediction results corresponding to each training corpus; calculate the loss between the prediction results and the labeled data corresponding to each training corpus, optimize the model parameters of the large language model according to the loss corresponding to each training corpus, and obtain the task model corresponding to the analysis instruction.

[0059] In some embodiments, the analysis and training module 303 is also configured as various analysis instructions, including: object attribute information extraction instructions, object user comment information mining instructions, object price trend analysis instructions and object development prospects analysis instructions; according to the object attribute information extraction instructions, the incrementally trained large language model is trained using multiple training corpora to obtain the task model corresponding to the object attribute information extraction instructions; according to the object user comment information mining instructions, the incrementally trained large language model is trained using multiple training corpora to obtain the task model corresponding to the object user comment information mining instructions; according to the object price trend analysis instructions, the incrementally trained large language model is trained using multiple training corpora to obtain the task model corresponding to the object price trend analysis instructions; according to the object development prospects analysis instructions, the incrementally trained large language model is trained using multiple training corpora to obtain the task model corresponding to the object development prospects analysis instructions.

[0060] In some embodiments, the collaborative analysis module 305 is also configured to process the text corpus in parallel through the task models corresponding to each analysis instruction to obtain the analysis results corresponding to each analysis instruction, wherein each task model exchanges analysis results with each other at a preset frequency during the process of processing the text corpus, and uses the analysis results of other task models as reference information to update its own analysis results; and summarizes the analysis results corresponding to each analysis instruction to form a comprehensive analysis report on the target object.

[0061] Furthermore, the large language model includes a vocabulary network, which is used to convert the input of the large language model into a specific sequence. Other networks in the large language model process the sequence output by the vocabulary network to obtain the output of the large language model; within the vocabulary network: the input of the large language model is divided into a sequence consisting of single words; the frequency of occurrence of each word is counted, the words that appear most frequently are merged into a new symbol, and the sequence of the words that appear most frequently is updated using the new symbol; through continuous iteration until the number of mergers reaches a preset number, the merged sequence is used as the sequence output by the vocabulary network, wherein each iteration merges the symbols that appear most frequently and updates the sequence of the symbols that appear most frequently.

[0062] In some embodiments, the collaborative analysis module 305 is also configured to process the text corpus through the task model corresponding to the object attribute information extraction instruction to obtain the analysis results corresponding to the object attribute information extraction instruction; obtain the analysis results corresponding to the object user comment information mining instruction through the task model corresponding to the object user comment information mining instruction; obtain the analysis results corresponding to the object price trend analysis instruction through the task model corresponding to the object price trend analysis instruction; obtain the analysis results corresponding to the object development prospect analysis instruction through the task model corresponding to the object development prospect analysis instruction; and summarize the analysis results corresponding to each analysis instruction to form a comprehensive analysis report on the target object.

[0063] It should be understood that the order of execution of each step in the above embodiment does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiment of the present disclosure.

[0064] Figure 4 FIG. 4 is a schematic diagram of an electronic device 4 provided in an embodiment of the present disclosure. Figure 4 As shown, electronic device 4 in this embodiment includes: a processor 401, a memory 402, and a computer program 403 stored in memory 402 and executable by processor 401. When processor 401 executes computer program 403, steps in the aforementioned method embodiments are implemented. Alternatively, when processor 401 executes computer program 403, the functions of the modules / units in the aforementioned device embodiments are implemented.

[0065] The electronic device 4 may be a desktop computer, a notebook, a PDA, a cloud server, or other electronic device. The electronic device 4 may include but is not limited to a processor 401 and a memory 402. Those skilled in the art will appreciate that Figure 4 The electronic device 4 is merely an example and does not limit the electronic device 4 , and may include more or fewer components than shown in the figure, or different components.

[0066] The processor 401 may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc.

[0067] Memory 402 can be an internal storage unit of electronic device 4, such as a hard drive or memory of electronic device 4. Memory 402 can also be an external storage device of electronic device 4, such as a plug-in hard drive, a Smart Media Card (SMC), a Secure Digital (SD) card, a flash memory card, etc. Memory 402 can also include both internal storage units of electronic device 4 and external storage devices. Memory 402 is used to store computer programs and other programs and data required by the electronic device.

[0068] Those skilled in the art will clearly understand that for the sake of convenience and brevity of description, only the above-mentioned functional units and module divisions are used as examples for illustration. In actual applications, the above-mentioned functions can be distributed to different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above. In the embodiments, each functional unit and module can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The above-mentioned integrated unit can be implemented in the form of hardware or in the form of software functional units.

[0069] If the integrated module / unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the present disclosure implements all or part of the processes in the above-mentioned embodiment method, and can also be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a computer-readable storage medium, and when the computer program is executed by the processor, it can implement the steps of the above-mentioned various method embodiments. The computer program may include computer program code, which may be in source code form, object code form, executable file or some intermediate form. The computer-readable medium may include: any entity or device capable of carrying computer program code, recording medium, USB flash drive, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electric carrier signal, telecommunication signal and software distribution medium, etc. It should be noted that the content contained in the computer-readable medium can be appropriately increased or decreased according to the legislation and patent practice requirements in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, computer-readable media do not include electric carrier signals and telecommunication signals.

[0070] The above embodiments are only used to illustrate the technical solutions of the present disclosure, rather than to limit them. Although the present disclosure has been described in detail with reference to the aforementioned embodiments, it should be understood by those skilled in the art that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features thereof may be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present disclosure, and should all be included in the protection scope of the present disclosure.

Claims

1. A method for generating a target object analysis report, characterized in that: include: Obtaining a training corpus, wherein the training corpus includes a plurality of training corpora related to different objects; Incrementally training a pre-trained large language model using multiple training corpora, so that the large language model learns relevant knowledge about the object corresponding to each training corpus; Generate multiple analysis instructions, and train the incrementally trained large language model using multiple training corpora according to each analysis instruction to obtain a task model corresponding to each analysis instruction; When the target object needs to be analyzed, text corpus related to the target object is retrieved; The text corpus is collaboratively processed through task models corresponding to various analysis instructions to form a comprehensive analysis report on the target object; the various analysis instructions include: object attribute information extraction instructions, object user comment information mining instructions, object price trend analysis instructions, and object development prospect analysis instructions; The text corpus is collaboratively processed by the task models corresponding to the respective analysis instructions to form a comprehensive analysis report on the target object, including: The text corpus is processed in parallel by the task models corresponding to the respective analysis instructions to obtain analysis results corresponding to the respective analysis instructions, wherein, in the process of processing the text corpus, the task models exchange analysis results with each other at a preset frequency and use the analysis results of other task models as reference information to update their own analysis results; The analysis results corresponding to each analysis instruction are aggregated to form a comprehensive analysis report on the target object.

2. The method according to claim 1, characterized in that Incrementally train the pre-trained large language model using multiple training corpora, so that the large language model learns relevant knowledge about the objects corresponding to each training corpus, including: The incremental training is contextual training, and the large language model predicts the subsequent words of a word or multiple consecutive words based on a word or multiple consecutive words in each training corpus; The large language model learns relevant knowledge about the object corresponding to each training corpus through the incremental training.

3. The method according to claim 1, characterized in that According to each analysis instruction, the incrementally trained large language model is trained using multiple training corpora to obtain a task model corresponding to each analysis instruction, including: For each analysis instruction: Construct the annotated data under the analysis instruction for each training corpus; The analysis instruction is used as a prompt word, the prompt word and each training corpus are input into the large language model after the incremental training, and the prediction result corresponding to each training corpus is output; The loss between the prediction result and the labeled data corresponding to each training corpus is calculated, and the model parameters of the large language model are optimized according to the loss corresponding to each training corpus to obtain the task model corresponding to the analysis instruction.

4. The method according to claim 1, characterized in that According to each analysis instruction, the incrementally trained large language model is trained using multiple training corpora to obtain a task model corresponding to each analysis instruction, including: According to the object attribute information extraction instruction, the incrementally trained large language model is trained using a plurality of training corpora to obtain a task model corresponding to the object attribute information extraction instruction; According to the user comment information mining instruction of the object, the incrementally trained large language model is trained using multiple training corpora to obtain a task model corresponding to the user comment information mining instruction of the object; According to the object price trend analysis instruction, the incrementally trained large language model is trained using a plurality of training corpora to obtain a task model corresponding to the object price trend analysis instruction; According to the object development prospect analysis instruction, the incrementally trained large language model is trained using multiple training corpora to obtain a task model corresponding to the object development prospect analysis instruction.

5. The method according to claim 1, characterized in that: The large language model includes a vocabulary network, which is used to convert the input of the large language model into a specific sequence. Other networks in the large language model process the sequence output by the vocabulary network to obtain the output of the large language model; Inside the vocabulary network: Dividing the input of the large language model into sequences of individual words; Count the frequency of occurrence of each word, merge the words that appear the most times into a new symbol, and use the new symbol to update the sequence of the words that appear the most times; By continuously iterating until the number of merging reaches a preset number, the merged sequence is used as the sequence output by the vocabulary network, wherein each iteration merges the symbol with the most occurrences and updates the sequence of the symbol with the most occurrences.

6. The method according to claim 1, characterized in that The text corpus is collaboratively processed by the task models corresponding to the respective analysis instructions to form a comprehensive analysis report on the target object, including: Various analysis instructions, including: object attribute information extraction instructions, object user comment information mining instructions, object price trend analysis instructions and object development prospect analysis instructions; Processing the text corpus through the task model corresponding to the object attribute information extraction instruction to obtain an analysis result corresponding to the object attribute information extraction instruction; Obtaining analysis results corresponding to the user review information mining instructions for the object through a task model corresponding to the user review information mining instructions for the object; Obtaining an analysis result corresponding to the object price trend analysis instruction through a task model corresponding to the object price trend analysis instruction; Obtaining an analysis result corresponding to the object development prospect analysis instruction through a task model corresponding to the object development prospect analysis instruction; The analysis results corresponding to each analysis instruction are aggregated to form a comprehensive analysis report on the target object.

7. A device for generating a target object analysis report, characterized in that: include: An acquisition module is configured to acquire a training corpus, wherein the training corpus includes a plurality of training corpora related to different objects; An incremental training module is configured to incrementally train a pre-trained large language model using multiple training corpora, so that the large language model learns relevant knowledge about the object corresponding to each training corpus; An analysis and training module is configured to generate a plurality of analysis instructions, and train the incrementally trained large language model using a plurality of training corpora according to each analysis instruction, to obtain a task model corresponding to each analysis instruction; A retrieval module is configured to retrieve text corpus related to a target object when analysis of the target object is required; a collaborative analysis module configured to collaboratively process the text corpus through task models corresponding to respective analysis instructions to form a comprehensive analysis report on the target object; the respective analysis instructions include: an object attribute information extraction instruction, an object user comment information mining instruction, an object price trend analysis instruction, and an object development prospect analysis instruction; The text corpus is collaboratively processed by the task models corresponding to the respective analysis instructions to form a comprehensive analysis report on the target object, including: The text corpus is processed in parallel by the task models corresponding to the respective analysis instructions to obtain analysis results corresponding to the respective analysis instructions, wherein, in the process of processing the text corpus, the task models exchange analysis results with each other at a preset frequency and use the analysis results of other task models as reference information to update their own analysis results; The analysis results corresponding to each analysis instruction are aggregated to form a comprehensive analysis report on the target object.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 6 are implemented.

9. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

Images