System and method for commercial viability report generation by ai

Abstract

The present application belongs to the field of commercial application, and particularly relates to a system and method for automatically generating a commercial feasibility report by using AI, aiming at solving the problem of high dependence on manpower and low efficiency in the existing commercial feasibility evaluation process. The method comprises the following steps: obtaining original data from multiple heterogeneous data sources; extracting multiple technology instances, constructing a technology knowledge graph, and determining corresponding time series vectors; inputting the technology knowledge graph into a preset hybrid prediction model to output a predicted momentum score; generating a business problem vector and a technology description vector based on a pre-trained language model, and determining an opportunity matching score matrix; receiving a target technology field input by a user, determining a technology entity cluster, and combining the predicted momentum score, the opportunity matching score matrix, and a preset report template to generate a feasibility report. The method generates a feasibility report by constructing a dynamic knowledge graph and combining a graph neural network and a time series prediction model, and has a better effect.

Description

Technical Field

[0001] This invention belongs to the field of commercial applications, specifically relating to a system and method for automatically generating commercial feasibility reports for AI-based technologies. Background Technology

[0002] Technology commercialization is a crucial step in transforming laboratory results into market products and creating economic value. It aims to conduct an in-depth technical feasibility assessment of the technological prototype and produce a comprehensive commercialization feasibility report. This report needs to comprehensively analyze multiple dimensions, including the technology's maturity, market prospects, competitive landscape, intellectual property status, potential risks, and returns.

[0003] Currently, the generation of such reports mainly relies on the following two methods:

[0004] One approach is based on manual research and writing, where a team of industry experts, market analysts, and technical personnel manually collects, organizes, and analyzes information by searching patent databases, academic literature, market reports, and news articles, ultimately writing a report. This method suffers from extremely low efficiency, high costs, and long processing times. Furthermore, manual analysis struggles to comprehensively cover the massive and rapidly updating heterogeneous data sources, and the results heavily rely on personal experience, making them prone to subjective bias and omissions, resulting in a lack of comprehensiveness and objectivity in the report.

[0005] Secondly, there's the auxiliary approach of using traditional information retrieval and simple analysis tools, such as database search tools or basic data analysis software. While this can improve the speed of information acquisition, it's essentially a data listing rather than intelligent analysis. Therefore, it lacks deep reasoning capabilities, cannot automatically uncover the intrinsic connections between technologies, cannot predict technological development trends, and cannot accurately match technological characteristics with business opportunities. The resulting reports are often static, isolated, and fragmented pieces of information, lacking reliable causal logic.

[0006] In summary, existing commercial feasibility assessment processes are highly reliant on human labor and inefficient. Therefore, it is necessary to propose a solution that can automatically, quickly, intelligently, and deeply generate high-quality commercial feasibility reports. Summary of the Invention

[0007] To address the aforementioned problems in existing technologies, namely the high reliance on human labor and low efficiency in current commercialization feasibility assessment processes, this invention, in its first aspect, proposes a method for automatically generating technology commercialization feasibility reports using AI, the method comprising:

[0008] Obtain raw data from multiple heterogeneous data sources;

[0009] Based on the original data, multiple technical entities are extracted. Based on the relationships between the technical entities, a technical knowledge graph is constructed with the technical entities as nodes and the relationships as edges. The corresponding time series vector is determined based on the quantitative indicators of each technical entity.

[0010] The technology knowledge graph is input into a preset hybrid prediction model, which includes a graph neural network and a long short-term memory network. The graph neural network is used to aggregate the topological structure and attribute information of each technology entity and its neighboring nodes, and outputs node embedding representations. The long short-term memory network is used to receive the node embedding representations and their corresponding time series vectors, perform time-series prediction of the development potential of each technology entity, and output a predicted momentum score.

[0011] Based on the pre-trained language model, business question vectors and technical description vectors corresponding to each technical entity are generated respectively. The opportunity matching score matrix is ​​determined by combining the business question vectors and technical description vectors.

[0012] The system receives the target technology field input by the user, identifies a cluster of technology entities in the technology knowledge graph, and generates a feasibility report for each technology entity in the cluster by combining the predicted momentum score, the chance matching score matrix, and a preset report template.

[0013] In some preferred embodiments, the graph neural network adopts a graph attention network architecture and generates a node embedding representation containing topological context by adaptively aggregating the feature information of multi-hop neighboring nodes by calculating the attention weights between nodes.

[0014] The Long Short-Term Memory Network is a multivariate time series prediction model. It receives the node embedding representation as a static feature and combines it with the time series vector, and inputs them together into a gated recurrent unit. The gated recurrent unit captures long-term dependencies and fuses spatiotemporal features to output the predicted momentum score of each technical entity in a future preset time interval.

[0015] In some preferred embodiments, the target technical field received from the user is used to determine technical entity clusters in the technical knowledge graph. Combined with the predicted momentum score, the opportunity matching score matrix, and a preset report template, a feasibility report for each technical entity in the technical entity cluster is generated, including:

[0016] The system receives a target technology field specified by the user as query input, queries the corresponding technology entity cluster in the technology knowledge graph, and determines the evolution path of the technology entity cluster in the knowledge graph.

[0017] The predicted momentum scores of the technical entity clusters are sorted, and the opportunity matching score matrix is ​​used to select at least one business application scenario with the highest score.

[0018] The preset structured report generation engine is invoked to automatically fill the corresponding chapters of the report template with the predicted momentum scores of the selected technical entities, the evolution path, and the business application scenarios, thereby obtaining the feasibility report.

[0019] In some preferred embodiments, receiving a user-specified target technical field as query input and querying the corresponding technical entity cluster in the technical knowledge graph includes:

[0020] The keywords of the target technology field are encoded into query vectors using a pre-trained language model;

[0021] Calculate the similarity between the query vector and the embedded representations of all technical nodes in the knowledge graph, and select nodes with similarity exceeding a preset threshold to form an initial node set;

[0022] The initial node set is analyzed based on a preset community detection algorithm, and the corresponding subgraph structure is identified in the initial node set. The subgraph structure is then used as the technical entity cluster.

[0023] In some preferred embodiments, the method further includes:

[0024] Identify at least one key assertion in the feasibility report that meets the preset requirements;

[0025] For each key assertion in the feasibility report, the original data points supporting the key assertion are extracted by tracing back to the technology knowledge graph as a traceable evidence chain, and the traceable evidence chain is integrated into the feasibility report.

[0026] In some preferred embodiments, the opportunity matching score matrix is ​​used to characterize the degree of matching between each technical entity and a predefined business problem, and the process of generating the opportunity matching score matrix includes:

[0027] A pre-trained language model, finely tuned from domain technical texts and business analysis corpora, is used to map the description text of each technical entity into a technical description vector in a high-dimensional space, and to map the business question into a business question vector.

[0028] Calculate the cosine similarity between each technical description vector and all business question vectors to determine the initial similarity matrix;

[0029] The initial similarity matrix is ​​completed using a matrix factorization-based collaborative filtering algorithm to obtain the chance matching score matrix.

[0030] In some preferred embodiments, generating a feasibility report for each technical entity in the technical entity cluster by combining the predicted momentum score, the opportunity matching score matrix, and a preset report template includes:

[0031] Determine the chapter structure defined in the preset report template;

[0032] Obtain structured data, which includes attribute information, predicted momentum scores, and scores in the chance matching score matrix for each technical entity in the technical entity cluster.

[0033] Using a pre-built text generation model, with the chapter structure as a constraint and the structured data as an input, coherent paragraph text is generated, and the feasibility report is formed by integrating the paragraph text.

[0034] In some preferred embodiments, the method further includes:

[0035] The feasibility report is displayed based on a preset user interface, and user feedback data on the feasibility report is received.

[0036] Based on the feedback data, a reinforcement learning reward signal is constructed, and the parameters of the graph neural network and long short-term memory network in the hybrid prediction model are fine-tuned.

[0037] In a second aspect, the present invention proposes a system for automatically generating commercial feasibility reports for AI-based technologies, the system comprising:

[0038] The data acquisition module is used to acquire raw data from multiple heterogeneous data sources;

[0039] The data extraction module is used to extract multiple technical entities from the original data, construct a technical knowledge graph with the technical entities as nodes and the relationships as edges based on the relationships between the technical entities, and determine the corresponding time series vector based on the quantitative indicators of each technical entity.

[0040] The model prediction module is used to input the technology knowledge graph into a preset hybrid prediction model. The hybrid prediction model includes a graph neural network and a long short-term memory network. The graph neural network is used to aggregate the topological structure and attribute information of each technology entity and its neighboring nodes, and output node embedding representations. The long short-term memory network is used to receive the node embedding representations and their corresponding time series vectors, perform time-series prediction of the development potential of each technology entity, and output a predicted momentum score.

[0041] The matching score module is used to generate business question vectors and technical description vectors corresponding to each technical entity based on a pre-trained language model, and to determine the opportunity matching score matrix by combining the business question vectors and technical description vectors.

[0042] The query output module is used to receive the target technology field input by the user, to determine the technology entity cluster in the technology knowledge graph, and to generate a feasibility report for each technology entity in the technology entity cluster by combining the predicted momentum score, the chance matching score matrix and the preset report template.

[0043] The beneficial effects of this invention are:

[0044] The method proposed in this invention visualizes and structures the technical entities (such as specific algorithms, materials, patents, and R&D institutions) and their complex relationships (such as citations, cooperation, competition, and upstream and downstream) scattered in different data sources by constructing a dynamic knowledge graph, integrating graph neural networks and time-series prediction models, and introducing a business opportunity matching mechanism. Ultimately, it realizes the automatic generation of intelligent reports from raw data.

[0045] Based on this, the manual analysis process, which used to take weeks or even months, can be shortened to minutes or hours. This effectively avoids subjective bias and uncertainty caused by human factors, ensuring the consistency and repeatability of the analysis results. Users only need to input a target technical field, and the system can automatically complete all steps from data collection, knowledge building, trend prediction to report generation, reducing decision-making and time costs.

[0046] This invention provides structured output based on a preset template. The report content is rooted in the aforementioned analysis (knowledge graph, momentum score, matching score), and also supports users to specify target technical fields, making the report generation more flexible and targeted, and able to meet the specific needs of different users. Attached Figure Description

[0047] Other features, objects, and advantages of this application will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings:

[0048] Figure 1 This is a flowchart illustrating a system method for automatically generating a commercial feasibility report for AI technology, as proposed in an embodiment of the present invention.

[0049] Figure 2 This is a schematic diagram of the structure of a system for automatically generating commercial feasibility reports for AI technology, as proposed in an embodiment of the present invention.

[0050] Figure 3 This is a schematic diagram of the structure of a computer system used to implement the methods and system embodiments of this application. Detailed Implementation

[0051] The present application will now be described in further detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are for illustrative purposes only and are not intended to limit the invention. Furthermore, it should be noted that, for ease of description, only the parts relevant to the invention are shown in the accompanying drawings.

[0052] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. This application will now be described in detail with reference to the accompanying drawings and embodiments.

[0053] like Figure 1 As shown, the first embodiment of the present invention provides a method for automatically generating a commercial feasibility report for AI technology, comprising:

[0054] Step S10: Obtain raw data from multiple heterogeneous data sources;

[0055] In this embodiment, the heterogeneous data source needs to cover multi-dimensional information in the technical field, including but not limited to:

[0056] Structured data: such as corporate R&D databases (e.g., R&D investment, number of patents), patent office databases (e.g., Espacenet's patent cataloging projects), industry statistics platforms (e.g., Gartner's market size data), etc.

[0057] Semi-structured data: such as technical standard documents (e.g., IEEE standards), metadata of academic papers (e.g., titles, abstracts, citations in Web of Science), etc.

[0058] Unstructured data: such as technical blogs (e.g., Medium), full text of patents (e.g., specifications of Google Patents), news reports (e.g., industry analysis from TechCrunch), etc.

[0059] Specifically, for structured data, it can be retrieved periodically through specific API interfaces (such as the Patent Office's REST API) or ETL tools (such as Apache NiFi);

[0060] For semi-structured data, XPath / JSONPath parsers can be used to extract key fields (such as "inventor" and "IPC classification number" in a patent).

[0061] For unstructured data, text can be downloaded via web crawlers (which must comply with the robots.txt protocol) or extracted using cloud services (such as AWS Textract).

[0062] More specifically, the raw data in this embodiment needs to be preprocessed to remove noise, and this preprocessing includes, but is not limited to:

[0063] Deduplication: Identify duplicate data using hash fingerprints (such as MD5) or semantic similarity (such as cosine similarity of BERT embeddings);

[0064] Missing value handling: For a small number of missing values, use the mean / median to fill (numerical type) or the mode to fill (categorical type); for a large number of missing data (such as a company having no R&D investment records for 3 consecutive years), mark them as "unknown" and remove them;

[0065] Standardized format: Time fields should be standardized to ISO 8601 format (e.g., "2023-10-01"), and numerical fields should be standardized to a single unit (e.g., "R&D investment" should be changed from "ten thousand yuan" to "one hundred million yuan").

[0066] Multilingual processing: Non-English data (such as Chinese patents) is converted to English via translation APIs (such as DeepL) to ensure consistency in subsequent processing.

[0067] For example, the full text of a patent downloaded from Google Patents (unstructured) is extracted using OCR, and then fields such as "title," "applicant," and "claims" are extracted using regular expressions (semi-structured). Finally, it is merged with the JSON data from the patent office (structured) to form a raw dataset in a unified format.

[0068] Step S20: Based on the original data, extract multiple technical entities, construct a technical knowledge graph with the technical entities as nodes and the relationships as edges based on the relationships between the technical entities, and determine the corresponding time series vector based on the quantitative indicators of each technical entity.

[0069] Among them, technical entities are the core nodes of the knowledge graph, which need to be extracted using domain-specific natural language processing (NLP) techniques:

[0070] Entity type definition: Define entity categories based on technical fields (such as artificial intelligence, new energy), including: technical terms (such as "Transformer model", "solid-state battery"); technical methods (such as "transfer learning", "sol-gel method"); technical components (such as "GPU", "proton exchange membrane"); application scenarios (such as "intelligent driving", "energy storage power station").

[0071] In this embodiment, a pre-trained model (such as BERT-TECH for the technical field) is fine-tuned to identify technical entities in the text. For example, in the patent abstract "This invention proposes a multimodal sentiment analysis method based on Transformer", the NER model can extract the entities: "Transformer model" (technical term) and "multimodal sentiment analysis" (technical method).

[0072] In this embodiment, the association relationship is used to define the semantic connection between entities, which needs to be extracted through syntactic analysis and knowledge reasoning, including but not limited to:

[0073] Explicit relation extraction: Utilize dependency parsing (such as Stanford CoreNLP) to identify subject-verb-object structures and extract direct relations (such as "X belongs to a subclass of Y" or "X applies to Y").

[0074] Implicit relation reasoning: Learn low-dimensional vector representations of entities through knowledge graph embedding (such as the RotatE model), calculate semantic similarity between entities, and infer indirect relationships (such as convolutional neural networks and recurrent neural networks both belong to "deep learning" and have a common relationship).

[0075] In this embodiment, each technical entity needs to be associated with dynamically changing quantitative indicators to form a time series vector. Specifically, the process includes:

[0076] Indicator selection: Select key indicators based on the technical field, such as:

[0077] Technology maturity: number of patents (years), number of paper citations (years);

[0078] Commercialization potential: Market size (RMB 100 million / year), enterprise R&D investment ratio (%);

[0079] Academic influence: H-index, number of highly cited papers (per year).

[0080] Time series construction: Aggregate indicators by time window (e.g., 1 year) to form a vector ve(t) = [ve(t1), ve(t2), ..., ve(tn)], where e represents the technology entity and ti represents the time point. For example, the time series vector of "artificial intelligence" may include the number of patents, the number of paper citations, etc. from 2018 to 2023.

[0081] Step S30: Input the technology knowledge graph into a preset hybrid prediction model. The hybrid prediction model includes a graph neural network and a long short-term memory network. The graph neural network is used to aggregate the topological structure and attribute information of each technology entity and its neighboring nodes, and output node embedding representations. The long short-term memory network is used to receive the node embedding representations and their corresponding time series vectors, perform time-series prediction of the development potential of each technology entity, and output a predicted momentum score.

[0082] In this embodiment, the topological aggregation technique of graph neural networks (GNNs) is used to capture the topological structure and attribute information of entities, generating node embedding representations:

[0083] The input to the graph structure includes the adjacency matrix A (representing the relationships between entities) and the node attribute matrix X (containing the static attributes of entities, such as the technology field, inventors, etc.) of the technology knowledge graph.

[0084] For the structural design of graph neural networks, the GraphSAGE model (suitable for large-scale graphs) can be adopted, which aggregates information of neighboring nodes through neighborhood sampling.

[0085] Specifically, for a node ei, its embedding representation hi(l+1) consists of its own attribute xi and the embeddings of its neighboring nodes. Weighted aggregation yields the following result, which satisfies the formula:

[0086] ;

[0087] Where W is the learnable weight matrix, Here, attention coefficients are used (computed via softmax), N(i) is the set of neighbors of i, and the final output is the embedding representation h of each technical entity ei. i (The dimensions are usually 128-256), which integrates topological structure and attribute information.

[0088] In this embodiment, a long short-term memory network is used for time series prediction:

[0089] LSTM receiver node embedding h i Using the time series vector ve(t), predict the development potential of a technological entity:

[0090] First, h i It interacts with the time series vector ve(t) or through a gating mechanism;

[0091] Next, the LSTM layer processes the historical information of the time series to capture trends and periodicity. For example, the input is h from the past 5 years. i Combined with the time series vector, it outputs the predicted momentum score for the next year;

[0092] Finally, the hidden states of the LSTM are mapped to predicted scores (ranging from 0 to 1, with 1 indicating high potential) through a fully connected layer (FC).

[0093] For example, for the "Transformer model," the GNN output embedding hi contains its association information with "natural language processing" and "computer vision," and the time series vector ve(t) contains the number of patents from 2018 to 2023. LSTM combines the two and predicts a development potential score of 0.85 (high potential) for 2024.

[0094] Step S40: Based on the pre-trained language model, generate business question vectors and technical description vectors corresponding to each technical entity, and combine the business question vectors and technical description vectors to determine the opportunity matching score matrix.

[0095] In this embodiment, a pre-trained language model (such as BERT) is used to generate vectors of business questions and technical descriptions:

[0096] For the business question vector, the vector is obtained by BERT encoding based on the business needs described by the input user (such as "how to reduce the battery cost of electric vehicles").

[0097] For the technical description vector, the vector is obtained by BERT encoding based on the technical document of the input technical entity (such as "solid-state batteries improve safety by replacing liquid electrolytes with solid electrolytes").

[0098] In this embodiment, the opportunity matching score matrix M is used to reflect the degree of matching between the technical entity and the business problem. The calculation process includes:

[0099] First, the similarity S1 between the business question vector and the technical description vector is calculated using cosine similarity.

[0100] Then, combining dimensions such as technological maturity and market demand, additional similarity is introduced (such as the matching degree between market size and business problem needs S2, and the matching degree between technological barriers and business problem risks S3).

[0101] Finally, calculate the elements of matrix M: M(e,p) = W1S1 + W2S2 + W3S3;

[0102] Where W is the weight, and the sum of the weights W1, W2, and W3 is 1, representing the matching score between the technical entity and the business problem.

[0103] For example, the cosine similarity between the vector of the business problem "reducing the cost of electric vehicle batteries" and the vector of the technological entity "solid-state batteries" is S1=0.82, the market size matching degree is S2=0.78, the technological barrier matching degree is S3=0.65, and the overall score is M=0.3×0.82+0.4×0.78+0.3×0.65=0.76 (high matching).

[0104] Step S50: Receive the target technology field input by the user to determine the technology entity cluster in the technology knowledge graph, and generate a feasibility report for each technology entity in the technology entity cluster by combining the predicted momentum score, the chance matching score matrix and the preset report template.

[0105] In this embodiment, after the user inputs a target technology field (such as "autonomous driving"), a community discovery algorithm (such as the Louvain algorithm) identifies closely related entity clusters in the knowledge graph. For example, the "autonomous driving" cluster may include entities such as "LiDAR", "computer vision", "path planning", and "V2X communication".

[0106] Specifically, the graph neural network adopts a graph attention network architecture and generates a node embedding representation containing topological context by calculating the attention weights between nodes to adaptively aggregate the feature information of multi-hop neighboring nodes.

[0107] The Long Short-Term Memory Network is a multivariate time series prediction model. It receives the node embedding representation as a static feature and combines it with the time series vector, and inputs them together into a gated recurrent unit. The gated recurrent unit captures long-term dependencies and fuses spatiotemporal features to output the predicted momentum score of each technical entity in a future preset time interval.

[0108] More specifically, the target technical field received from the user is used to determine technical entity clusters in the technical knowledge graph. Combining the predicted momentum score, the opportunity matching score matrix, and a preset report template, a feasibility report for each technical entity in the technical entity cluster is generated, including:

[0109] The system receives a target technology field specified by the user as query input, queries the corresponding technology entity cluster in the technology knowledge graph, and determines the evolution path of the technology entity cluster in the knowledge graph.

[0110] The predicted momentum scores of the technical entity clusters are sorted, and the opportunity matching score matrix is ​​used to select at least one business application scenario with the highest score.

[0111] The preset structured report generation engine is invoked to automatically fill the corresponding chapters of the report template with the predicted momentum scores of the selected technical entities, the evolution path, and the business application scenarios, thereby obtaining the feasibility report.

[0112] More specifically, receiving a user-specified target technical field as query input and querying the corresponding technical entity cluster in the technical knowledge graph includes:

[0113] The keywords of the target technology field are encoded into query vectors using a pre-trained language model;

[0114] Calculate the similarity between the query vector and the embedded representations of all technical nodes in the knowledge graph, and select nodes with similarity exceeding a preset threshold to form an initial node set;

[0115] The initial node set is analyzed based on a preset community detection algorithm, and the corresponding subgraph structure is identified in the initial node set. The subgraph structure is then used as the technical entity cluster.

[0116] More specifically, the method further includes:

[0117] Identify at least one key assertion in the feasibility report that meets the preset requirements;

[0118] For each key assertion in the feasibility report, the original data points supporting the key assertion are extracted by tracing back to the technology knowledge graph as a traceable evidence chain, and the traceable evidence chain is integrated into the feasibility report.

[0119] More specifically, the opportunity matching score matrix is ​​used to characterize the degree of matching between each technical entity and a predefined business problem, and the process of generating the opportunity matching score matrix includes:

[0120] A pre-trained language model, finely tuned from domain technical texts and business analysis corpora, is used to map the description text of each technical entity into a technical description vector in a high-dimensional space, and to map the business question into a business question vector.

[0121] Calculate the cosine similarity between each technical description vector and all business question vectors to determine the initial similarity matrix;

[0122] The initial similarity matrix is ​​completed using a matrix factorization-based collaborative filtering algorithm to obtain the chance matching score matrix.

[0123] More specifically, the step of generating a feasibility report for each technical entity in the technical entity cluster by combining the predicted momentum score, the opportunity matching score matrix, and a preset report template includes:

[0124] Determine the chapter structure defined in the preset report template;

[0125] Obtain structured data, which includes attribute information, predicted momentum scores, and scores in the chance matching score matrix for each technical entity in the technical entity cluster.

[0126] Using a pre-built text generation model, with the chapter structure as a constraint and the structured data as an input, coherent paragraph text is generated, and the feasibility report is formed by integrating the paragraph text.

[0127] More specifically, the method further includes:

[0128] The feasibility report is displayed based on a preset user interface, and user feedback data on the feasibility report is received.

[0129] Based on the feedback data, a reinforcement learning reward signal is constructed, and the parameters of the graph neural network and long short-term memory network in the hybrid prediction model are fine-tuned.

[0130] Although the steps in the above embodiments are described in the above order, those skilled in the art will understand that in order to achieve the effect of this embodiment, different steps do not need to be executed in such an order. They can be executed simultaneously (in parallel) or in a reverse order. These simple variations are all within the protection scope of this invention.

[0131] The second embodiment of the present invention proposes a system for automatically generating commercial feasibility reports for AI technology, which is used to implement the method involved in the first embodiment above.

[0132] The system for automatically generating commercial feasibility reports for AI technology can be a software module. This software module includes several instructions stored in memory. The processor can access this memory, call the instructions, and execute them to complete the system method for automatically generating commercial feasibility reports for AI technology as described in the various embodiments above. In some embodiments, the system for automatically generating commercial feasibility reports for AI technology can also be built from hardware devices. For example, the system can be built from one or more chips, which can work in coordination to complete the system method for automatically generating commercial feasibility reports for AI technology as described in the various embodiments above. Furthermore, the system can also be built from various logic devices, such as general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), microcontrollers, ARM (AcornRISC) or other programmable logic devices, discrete gate or transistor logic, discrete hardware components, or any combination of these components.

[0133] Please refer to Figure 2 , Figure 2 The diagram shows the structure of the system for automatically generating commercial feasibility reports for this AI technology. As shown in the figure, the system includes:

[0134] The data acquisition module 210 is used to acquire raw data from multiple heterogeneous data sources;

[0135] The data extraction module 220 is used to extract multiple technical entities based on the original data, construct a technical knowledge graph with the technical entities as nodes and the relationships as edges based on the relationships between the technical entities, and determine the corresponding time series vector based on the quantitative indicators of each technical entity.

[0136] The model prediction module 230 is used to input the technology knowledge graph into a preset hybrid prediction model. The hybrid prediction model includes a graph neural network and a long short-term memory network. The graph neural network is used to aggregate the topological structure and attribute information of each technology entity and its neighboring nodes, and output node embedding representations. The long short-term memory network is used to receive the node embedding representations and their corresponding time series vectors, perform time-series prediction of the development potential of each technology entity, and output a predicted momentum score.

[0137] The matching score module 240 is used to generate business question vectors and technical description vectors corresponding to each technical entity based on a pre-trained language model, and to determine the opportunity matching score matrix by combining the business question vectors and technical description vectors.

[0138] The query output module 250 is used to receive the target technology field input by the user, to determine the technology entity cluster in the technology knowledge graph, and to generate a feasibility report for each technology entity in the technology entity cluster by combining the predicted momentum score, the chance matching score matrix and the preset report template.

[0139] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working process and related descriptions of the system described above can be found in the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0140] It should be noted that the AI-automated technology commercialization feasibility report system provided in the above embodiments is only an example of the division of the above functional modules. In practical applications, the above functions can be assigned to different functional modules as needed, that is, the modules or steps in the embodiments of the present invention can be further decomposed or combined. For example, the modules in the above embodiments can be merged into one module, or further divided into multiple sub-modules to complete all or part of the functions described above. The names of the modules and steps involved in the embodiments of the present invention are only for distinguishing the various modules or steps and are not considered as an improper limitation of the present invention.

[0141] The third embodiment of the present invention provides a computer-readable storage medium storing computer instructions, which are executed by the computer to implement the above-described system method for automatically generating a commercial feasibility report for AI technology.

[0142] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working process and related descriptions of the electronic device and computer-readable storage medium described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0143] Those skilled in the art will recognize that the modules and method steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. The programs corresponding to the software modules and method steps can be placed in random access memory (RAM), main memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, hard disks, removable disks, CD-ROMs, or any other form of storage medium known in the art. To clearly illustrate the interchangeability of electronic hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in electronic hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of the invention.

[0144] The following is for reference. Figure 3 It shows a schematic diagram of the structure of a computer system used to implement the methods and system embodiments of this application. Figure 3 The server shown is merely an example and should not impose any limitations on the functionality and scope of use of the embodiments of this application.

[0145] like Figure 3 As shown, the computer system includes a Central Processing Unit (CPU) 301, which can perform various appropriate actions and processes based on programs stored in Read Only Memory (ROM) 302 or programs loaded from storage section 308 into Random Access Memory (RAM) 303. The RAM 303 also stores various programs and data required for system operation. The CPU 301, ROM 302, and RAM 303 are interconnected via a bus 304. An Input / Output (I / O) interface 305 is also connected to the bus 304.

[0146] The following components are connected to the input / output interface 305: an input section 306 including a keyboard, mouse, etc.; an output section 307 including a cathode ray tube (CRT), liquid crystal display (LCD), etc., and speakers, etc.; a storage section 308 including a hard disk, etc.; and a communication section 309 including a network interface card such as a LAN (Local Area Network) card, modem, etc. The communication section 309 performs communication processing via a network such as the Internet. A drive 310 is also connected to the input / output interface 305 as needed. A removable medium 311, such as a disk, optical disk, magneto-optical disk, semiconductor memory, etc., is installed on the drive 310 as needed so that computer programs read from it can be installed into the storage section 308 as needed.

[0147] Specifically, according to embodiments of this disclosure, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of this disclosure include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via communication section 309, and / or installed from removable medium 311. When the computer program is executed by central processing unit (CPU) 301, it performs the functions defined in the methods of this application. It should be noted that the computer-readable medium described above in this application can be a computer-readable signal medium or a computer-readable storage medium, or any combination of the two. The computer-readable storage medium can be, for example,—but not limited to—an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, system, or device, or any combination thereof. More specific examples of computer-readable storage media may include, but are not limited to: electrical connections having one or more wires, portable computer disks, hard disks, random access memory, read-only memory, erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory, optical storage devices, magnetic storage devices, or any suitable combination thereof. In this application, a computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in connection with an instruction execution system, system, or device. In this application, a computer-readable signal medium may include a data signal propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals can take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. A computer-readable signal medium can also be any computer-readable medium other than a computer-readable storage medium, which can send, propagate, or transmit a program for use by or in connection with an instruction execution system, system, or device. The program code contained on the computer-readable medium can be transmitted using any suitable medium, including but not limited to: wireless, wire, optical fiber, RF, etc., or any suitable combination thereof.

[0148] Computer program code for performing the operations of this application can be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, and C++, and conventional procedural programming languages ​​such as the "C" language or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0149] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0150] The terms “first”, “second”, etc., are used to distinguish similar objects, not to describe or indicate a specific order or sequence.

[0151] The term "comprising" or any other similar term is intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus / system that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to those processes, methods, articles, or apparatus / systems.

[0152] The technical solution of the present invention has been described above with reference to the preferred embodiments shown in the accompanying drawings. However, it will be readily understood by those skilled in the art that the scope of protection of the present invention is obviously not limited to these specific embodiments. Without departing from the principles of the present invention, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after these changes or substitutions will all fall within the scope of protection of the present invention.

Claims

1. A method for AI automatically generating a commercial feasibility report, characterized in that, The method comprises: acquiring raw data from a plurality of heterogeneous data sources; extracting a plurality of technical entities from the raw data, constructing a technical knowledge graph with the technical entities as nodes and the association relationships between the technical entities as edges, and determining a time series vector for each technical entity according to a quantitative indicator of each technical entity; inputting the technical knowledge graph into a preset hybrid prediction model, the hybrid prediction model comprising a graph neural network and a long short-term memory network, the graph neural network being configured to aggregate topological structure and attribute information of each technical entity and its adjacent nodes to output a node embedding representation, and the long short-term memory network being configured to receive the node embedding representation and its corresponding time series vector, perform time series prediction on the development potential of each technical entity, and output a predicted momentum score; generating a business problem vector and a technical description vector corresponding to each technical entity based on a pre-trained language model, and determining an opportunity matching score matrix by combining the business problem vector and the technical description vector; receiving a target technical field input by a user, determining a technical entity cluster in the technical knowledge graph, and generating a feasibility report of each technical entity in the technical entity cluster by combining the predicted momentum score, the opportunity matching score matrix, and a preset report template; wherein the receiving of the target technical field input by the user, the determination of the technical entity cluster in the technical knowledge graph, and the generation of the feasibility report of each technical entity in the technical entity cluster by combining the predicted momentum score, the opportunity matching score matrix, and the preset report template comprise: receiving a target technical field specified by a user as a query input, querying a corresponding technical entity cluster in the technical knowledge graph, and determining an evolution path of the technical entity cluster in the knowledge graph; sorting according to the predicted momentum score of the technical entity cluster, and screening at least one business application scenario with the highest score by combining the opportunity matching score matrix; calling a preset structured report generation engine to automatically fill the predicted momentum score of the screened technical entity, the evolution path, and the business application scenario into corresponding chapters of the report template to obtain the feasibility report. 2.The method of commercializing an AI automatically generated technical feasibility report according to claim 1, wherein, The graph neural network adopts a graph attention network architecture and adaptively aggregates feature information of multi-hop neighbor nodes by calculating attention weights between nodes to generate a node embedding representation containing topological context. The long short-term memory network is a multivariate time series prediction model, receives the node embedding representation as a static feature, and combines the time series vector to input into a gated recurrent unit, which outputs a predicted momentum score of each technical entity in a future preset time interval by capturing long-term dependencies and fusing spatio-temporal features. 3.The method of commercializing an AI automatically generated technical feasibility report according to claim 1, wherein, The receiving of the target technical field specified by the user as the query input and the querying of the corresponding technical entity cluster in the technical knowledge graph comprise: encoding keywords of the target technical field into a query vector through a pre-trained language model. Calculate the similarity of the query vector and the embedding representation of all technology nodes in the knowledge graph, and select nodes with similarity exceeding a preset threshold to form an initial node set; Based on a preset community discovery algorithm, analyze the initial node set, identify the corresponding subgraph structure in the initial node set, and take the subgraph structure as the technical entity cluster. 4.The method of commercializing an AI automatically generated technical feasibility report according to claim 1, wherein, The method further comprises: Determine at least one key argument in the feasibility report that meets the preset requirements; For each key argument in the feasibility report, extract the original data points used to support the key argument as a trace evidence chain by backtracking to the technical knowledge graph, and integrate the trace evidence chain into the feasibility report. 5.The method of commercializing AI automatically generated technical feasibility reports of claim 1, wherein, The opportunity matching score matrix is used to represent the matching degree between each technical entity and a predefined business problem. The generation process of the opportunity matching score matrix includes: Using a pre-trained language model fine-tuned on domain technical text and business analysis corpus, respectively mapping the description text of each technical entity into a technical description vector in a high-dimensional space, and mapping the business problem into a business problem vector; Calculate the cosine similarity between each technical description vector and all business problem vectors to determine an initial similarity matrix; Use a matrix decomposition-based collaborative filtering algorithm to complete the initial similarity matrix to obtain the opportunity matching score matrix. 6.The method of commercializing an AI automatically generated technical feasibility report according to claim 1, wherein, The combination of the predicted momentum score, the opportunity matching score matrix, and the preset report template to generate the feasibility report of each technical entity in the technical entity cluster includes: Determine the chapter structure defined in the preset report template; Obtain structured data, including attribute information of each technical entity in the technical entity cluster, predicted momentum score, and score in the opportunity matching score matrix; Use a pre-built text generation model to generate coherent paragraph text with the chapter structure as a constraint and the structured data as an input condition, and form the feasibility report by integrating the paragraph text.

7. The method of claim 1-6, wherein, The method further comprises: Display the feasibility report based on a preset user interface and receive user feedback data on the feasibility report; Construct a reinforcement learning reward signal based on the feedback data to fine-tune the parameters of the graph neural network and the long short-term memory network in the hybrid prediction model.

8. A system for AI automatically generating a commercial feasibility report, characterized by, The system comprises: A data acquisition module for acquiring raw data from multiple heterogeneous data sources; A data extraction module for extracting multiple technical entities from the raw data, constructing a technical knowledge graph with the technical entities as nodes and the association relationships as edges based on the association relationships between the technical entities, and determining a corresponding time series vector for each technical entity based on a quantitative indicator. The model prediction module is configured to input the technical knowledge graph into a preset hybrid prediction model, the hybrid prediction model comprising a graph neural network and a long short-term memory network, the graph neural network being configured to aggregate topological structure and attribute information of each technical entity and its adjacent nodes, and output node embedding representation; the long short-term memory network being configured to receive the node embedding representation and its corresponding time sequence vector, perform time series prediction on development potential of each technical entity, and output a predicted momentum score; The matching score module is configured to generate a business problem vector and a technical description vector corresponding to each technical entity based on a pre-trained language model, and determine an opportunity matching score matrix by combining the business problem vector and the technical description vector; The query output module is configured to receive a target technical field input by a user, determine a technical entity cluster in the technical knowledge graph, and generate a feasibility report of each technical entity in the technical entity cluster by combining the predicted momentum score, the opportunity matching score matrix, and a preset report template. The receiving of the target technical field input by the user, the determination of the technical entity cluster in the technical knowledge graph, and the generation of the feasibility report of each technical entity in the technical entity cluster by combining the predicted momentum score, the opportunity matching score matrix, and the preset report template comprise: Receiving a target technical field specified by a user as a query input, querying a corresponding technical entity cluster in the technical knowledge graph, and determining an evolution path of the technical entity cluster in the knowledge graph; Sorting according to the predicted momentum score of the technical entity cluster, and screening at least one business application scenario with the highest score in combination with the opportunity matching score matrix; Calling a preset structured report generation engine, and automatically filling the predicted momentum score of the screened technical entity, the evolution path, and the business application scenario into corresponding chapters of the report template to obtain the feasibility report.

Images