Information processing method and device, electronic equipment and storage medium
The pre-trained big model technology generates address inference logic and code, combined with multimodal retrieval, solves the problem of identifying address information diversity in banking financial services, realizes accurate and clear description of address information, and reduces human errors.
Patent Information
- Application Number
- CN202411169126.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-23
- Publication Date
- 2024-11-26
AI Technical Summary
In banking and financial business, the diversity of address information and differences in expression habits make it difficult to ensure the consistency and accuracy of data, and it is difficult for the prior art to effectively identify and process address information entered by users.
Using pre-trained big model technology, by obtaining the task request information input by the user, determining the prompt words of the target big model, generating the graph structure of the address inference logic, and executing the address inference code to obtain text address information, combining multimodal retrieval and Feature extraction improves the accuracy and clarity of address information.
It realizes automated inference and identification of address information, reduces human errors, improves the accuracy and clarity of address information, and enhances its practicality.
Smart Images

Figure CN119026690A_ABST
Abstract
Description
Technical Field
[0001] The present disclosure relates to the fields of artificial intelligence and large model technology, and in particular to an information processing method, device, electronic device and storage medium. Background Art
[0002] In banking and financial services, address information is one of the indispensable customer attributes, covering information such as household registration address, work address, credit card bill address, etc. However, due to different expression habits of different people, the same address may exist in many different forms, which involves omissions, abbreviations, typos, etc., which brings difficulties to the consistency and accuracy of data. Summary of the invention
[0003] The present disclosure provides an information processing method, device, electronic device and storage medium, which are used to solve the problem that the address of information input by the user cannot be accurately recognized.
[0004] To this end, one objective of the present disclosure is to provide an information processing method.
[0005] A second objective of the present disclosure is to provide an information processing device.
[0006] A third objective of the present disclosure is to provide an electronic device.
[0007] A fourth object of the present disclosure is to provide a non-transitory computer-readable storage medium.
[0008] A fifth object of the present disclosure is to provide a computer program product.
[0009] To achieve the above-mentioned purpose, the first aspect of the present disclosure proposes an information processing method, including: obtaining task request information input by a user; based on the task request information, determining a first prompt word of a pre-trained target big model, and inputting the first prompt word into the target big model; the target big model determines a graph structure of an address reasoning logic based on the first prompt word, and based on the graph structure of the address reasoning logic, generates an address reasoning code, and executes the address reasoning code to obtain first text address information associated with the task request information; performing information retrieval on the first text address information, determining a multimodal retrieval result of the first text address information, and determining a first spatial description information of the first text address information based on the multimodal retrieval result.
[0010] According to one embodiment of the present disclosure, determining the first prompt word of the pre-trained target large model based on the task request information includes: obtaining an initial second prompt word based on the task request information and according to a prompt word generation strategy; and supplementing the second prompt word based on a set reasoning mode of the target large model to obtain the first prompt word of the target large model.
[0011] According to one embodiment of the present disclosure, the set reasoning mode based on the target large model supplements the second prompt word to obtain the first prompt word of the large model, including: obtaining node information of each reasoning step in the set reasoning mode, the node information at least including a first node and a second node; connecting each reasoning step in the set reasoning mode based on the first node and the second node to obtain a spliced reasoning step; determining a third prompt word based on the spliced reasoning step, and splicing the third prompt word and the second prompt word to obtain the first prompt word.
[0012] According to one embodiment of the present disclosure, the address reasoning code is generated based on the graph structure of the address reasoning logic, including: obtaining the code constraints of the target large model configuration, and performing code reasoning on the graph structure of the address reasoning logic based on the code constraints to generate the address reasoning code.
[0013] According to one embodiment of the present disclosure, determining the first prompt word of the pre-trained target large model based on the task request information also includes: performing intent recognition on the task request information to determine the task type; in response to the task type being a reasoning task, determining the first prompt word of the pre-trained target large model based on the task request information.
[0014] According to one embodiment of the present disclosure, the intent recognition of the task request information to determine the task type includes: determining invalid information in the task request information, and filtering the invalid information to obtain valid information in the task request information; identifying the valid information to determine whether the valid information carries address information; if the valid information does not carry address information, determining that the task type is a reasoning task.
[0015] According to one embodiment of the present disclosure, the method also includes: if the valid information carries address information, determining that the task type is the first type; segmenting the valid information to obtain second text address information associated with the task request information; and performing information retrieval on the second text address information to determine second spatial description information of the second text address information.
[0016] According to one embodiment of the present disclosure, segmenting the valid information to obtain the second address information includes: determining resource requirement information for information segmentation; in response to the resource requirement information being greater than a resource threshold, performing feature extraction on the valid information to obtain feature data of the valid information; and performing conditional random field (CRF) segmentation on the feature data to obtain the second address information.
[0017] According to one embodiment of the present disclosure, the method further includes: in response to the resource demand information being less than or equal to the resource threshold, using a neural network-based bidirectional encoder to represent a BERT model, adjusting the valid information to obtain adjusted valid information; performing CRF segmentation on the adjusted valid information to obtain the second address information.
[0018] According to one embodiment of the present disclosure, the information retrieval process includes: encoding any text address information in the first text address information and the second text address information to obtain a first coded representation; acquiring multimodal data stored in a database, performing feature extraction on the multimodal data to obtain multiple feature vectors of the multimodal data, and fusing the multiple feature vectors to obtain a second coded representation of the multimodal data; acquiring the similarity between the first coded representation and the second coded representation, and determining a multimodal retrieval result matching any text address information from the database based on the similarity.
[0019] According to one embodiment of the present disclosure, the method also includes: obtaining address elements of the first spatial description information and the second spatial description information, and obtaining points of interest of the first spatial description information and the second spatial description information; for any element of the address elements and the points of interest, determining the layers and connections between any of the elements; using any of the elements as entity information to constitute a node in the address space knowledge graph, using the layers and connections as the relationship between entities to constitute an edge in the address space knowledge graph, and connecting the nodes and the edges to constitute an address space knowledge graph.
[0020] To achieve the above-mentioned purpose, the second aspect of the present disclosure proposes an information processing device, including: an acquisition module, used to acquire task request information input by a user; a determination module, used to determine a first prompt word of a pre-trained target large model based on the task request information, and input the first prompt word into the target large model; an inference module, used to determine the graph structure of the address reasoning logic based on the first prompt word by the target large model, and generate an address reasoning code based on the graph structure of the address reasoning logic, and execute the address reasoning code to obtain the first text address information associated with the task request information; a retrieval module, used to perform information retrieval on the first text address information, determine a multimodal retrieval result of the first text address information, and determine the first spatial description information of the first text address information based on the multimodal retrieval result.
[0021] According to one embodiment of the present disclosure, the determination module is also used to: based on the task request information, obtain an initial second prompt word according to the prompt word generation strategy; based on the set reasoning mode of the target large model, supplement the second prompt word to obtain the first prompt word of the target large model.
[0022] According to one embodiment of the present disclosure, the determination module is also used to: obtain node information of each reasoning step in the set reasoning mode, and the node information includes at least a first node and a second node; connect each reasoning step in the set reasoning mode based on the first node and the second node to obtain a spliced reasoning step; determine a third prompt word based on the spliced reasoning step, and splice the third prompt word and the second prompt word to obtain the first prompt word.
[0023] According to one embodiment of the present disclosure, the reasoning module is further used to: obtain code constraints configured by the target large model, and based on the code constraints, perform code reasoning on the graph structure of the address reasoning logic to generate the address reasoning code.
[0024] According to one embodiment of the present disclosure, the determination module is further used to: perform intent recognition on the task request information to determine the task type; in response to the task type being a reasoning task, determine the first prompt word of the pre-trained target large model based on the task request information.
[0025] According to one embodiment of the present disclosure, the determination module is also used to: determine invalid information in the task request information, and filter the invalid information to obtain valid information in the task request information; identify the valid information to determine whether the valid information carries address information; if the valid information does not carry address information, determine that the task type is a reasoning task.
[0026] According to one embodiment of the present disclosure, the reasoning module is also used to: if the valid information carries address information, determine that the task type is the first type; segment the valid information to obtain second text address information associated with the task request information; and perform information retrieval on the second text address information to determine second spatial description information of the second text address information.
[0027] According to one embodiment of the present disclosure, the reasoning module is also used to: determine resource requirement information for information segmentation; in response to the resource requirement information being greater than a resource threshold, perform feature extraction on the valid information to obtain feature data of the valid information; perform conditional random field (CRF) segmentation on the feature data to obtain the second address information.
[0028] According to one embodiment of the present disclosure, the device also includes: in response to the resource demand information being less than or equal to the resource threshold, using a neural network-based bidirectional encoder to represent the BERT model, adjusting the valid information to obtain adjusted valid information; performing CRF segmentation on the adjusted valid information to obtain the second address information.
[0029] According to one embodiment of the present disclosure, the retrieval module is also used to: encode any text address information in the first text address information and the second text address information to obtain a first coded representation; obtain multimodal data stored in a database, perform feature extraction on the multimodal data to obtain multiple feature vectors of the multimodal data, and fuse the multiple feature vectors to obtain a second coded representation of the multimodal data; obtain the similarity between the first coded representation and the second coded representation, and determine a multimodal retrieval result matching any text address information from the database based on the similarity.
[0030] According to one embodiment of the present disclosure, the device also includes: obtaining address elements of the first spatial description information and the second spatial description information, and obtaining points of interest of the first spatial description information and the second spatial description information; for any element of the address elements and the points of interest, determining the layers and connections between any of the elements; using any of the elements as entity information to constitute a node in the address space knowledge graph, using the layers and connections as the relationship between entities to constitute an edge in the address space knowledge graph, and connecting the nodes and the edges to constitute an address space knowledge graph.
[0031] To achieve the above-mentioned purpose, the third aspect embodiment of the present disclosure proposes an electronic device, comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to implement the information processing method as described in the first aspect embodiment of the present disclosure.
[0032] To achieve the above-mentioned purpose, the fourth aspect embodiment of the present disclosure proposes a non-transitory computer-readable storage medium storing computer instructions, wherein the computer instructions are used to implement the information processing method as described in the first aspect embodiment of the present disclosure.
[0033] To achieve the above-mentioned purpose, the fifth aspect of the present disclosure proposes a computer program product, including a computer program, which is used to implement the information processing method as described in the first aspect of the present disclosure when executed by a processor. BRIEF DESCRIPTION OF THE DRAWINGS
[0034] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present disclosure and, together with the description, serve to explain the principles of the present disclosure.
[0035] Figure 1 An example flow chart of an information processing method provided by an embodiment of the present disclosure;
[0036] Figure 2 An example diagram of the graph structure of the address reasoning logic provided by the embodiment of the present disclosure;
[0037] Figure 3 An example flowchart of another information processing method provided by an embodiment of the present disclosure;
[0038] Figure 4 An example flowchart of another information processing method provided by an embodiment of the present disclosure;
[0039] Figure 5 An example flowchart of another information processing method provided by an embodiment of the present disclosure;
[0040] Figure 6 An example diagram of the address reasoning process for the reasoning task provided in the embodiment of the present disclosure;
[0041] Figure 7 An example flowchart of another information processing method provided by an embodiment of the present disclosure;
[0042] Figure 8 An example diagram of a process for address recognition of task request information provided in an embodiment of the present disclosure;
[0043] Fig. 9An exemplary structural diagram of an information processing device provided in an embodiment of the present disclosure;
[0044] Fig.10 This is a diagram showing an example structure of an electronic device provided in an embodiment of the present disclosure.
[0045] The above drawings have shown clear embodiments of the present disclosure, which will be described in more detail below. These drawings and text descriptions are not intended to limit the scope of the present disclosure in any way, but to illustrate the concepts of the present disclosure to those skilled in the art by referring to specific embodiments. DETAILED DESCRIPTION
[0046] In order to enable ordinary persons in the art to better understand the technical solutions of the present disclosure, the technical solutions in the embodiments of the present disclosure will be clearly and completely described below in conjunction with the accompanying drawings.
[0047] It should be noted that the terms "first", "second", etc. in the specification and claims of the present disclosure and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments of the present disclosure described herein can be implemented in an order other than those illustrated or described herein. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present disclosure. Instead, they are merely examples of devices and methods consistent with some aspects of the present disclosure as detailed in the appended claims.
[0048] It should be noted that the acquisition, storage, use, and processing of data in the technical solution of this application are in compliance with the provisions of relevant laws and regulations.
[0049] It should be noted that the information (including but not limited to user device information, user personal information, etc.), data (including but not limited to data used for analysis, stored data, displayed data, etc.) and signals involved in this application are all authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with relevant laws, regulations and standards.
[0050] It should be noted that in the embodiments of the present application, certain software, components, models and other existing solutions in the industry may be mentioned, and they should be regarded as exemplary. Their purpose is only to illustrate the feasibility of implementing the technical solution of the present application, but it does not mean that the applicant has or will necessarily use the relevant content of the solution.
[0051] Figure 1 This is an example flow chart of the information processing method provided by the embodiment of the present disclosure. Figure 1 As shown, the information processing method includes:
[0052] S101, obtaining task request information input by a user.
[0053] It should be noted that the execution subject of the information processing method provided in the embodiment of the present disclosure is an electronic device, and the electronic device may be a terminal device. Optionally, the terminal device may be a mobile electronic device or a non-mobile electronic device. Exemplarily, the mobile electronic device may be a mobile phone, a tablet computer, a laptop computer, a PDA, an in-vehicle electronic device, a wearable device, an ultra-mobile personal computer (UMPC), a netbook or a personal digital assistant (PDA), etc., and the non-mobile electronic device may be a network attached storage (NAS), a personal computer (PC), a television (TV), a teller machine or a self-service machine, etc. The embodiment of the present disclosure is not specifically limited.
[0054] In some implementations, the information input by the user at a specified position of the electronic device can be used as the task request information input by the user. Optionally, the user can input voice information, text information, etc. The embodiment of the present disclosure does not specifically limit the form of the user input information.
[0055] Optionally, when the information input by the user is voice information, the voice information can be converted into text information, and the text information is used as the task request information; when the information input by the user is text information, the input text information is the task request information.
[0056] S102, based on the task request information, determine the first prompt word of the pre-trained target large model, and input the first prompt word into the target large model.
[0057] In some implementations, the format of the task request information can be pre-set, where the set format includes prompt word information. That is, the user can input the task request information according to the set format, and then extract the first prompt word from the task request information, and input the first prompt word into the pre-trained target large model.
[0058] S103, the target large model determines the graph structure of the address reasoning logic based on the first prompt word, generates address reasoning code based on the graph structure of the address reasoning logic, and executes the address reasoning code to obtain the first text address information associated with the task request information.
[0059] In some implementations, the target big model can obtain the task description carried in the first prompt word by identifying the first prompt word, wherein the task description is used to indicate the task that the target big model needs to solve, and then a flowchart for solving the task can be generated, that is, a graph structure of the address reasoning logic, so as to decompose the process of reasoning address information into operational steps, thereby more accurately obtaining the first text address information.
[0060] For example, if the task request information is "need to open a store near a school, with a wide road and good greenery", the first prompt word is: near a school, with a wide road and good greenery. The target large model determines the logic and steps of address reasoning based on the first prompt word, such as determining the address with good greenery, then filtering the address with a wide road from the addresses with good greenery, and finally determining the address near the school from the addresses with a wide road. Then, based on the above address reasoning logic and steps, a graph structure of the address reasoning logic can be generated.
[0061] like Figure 2 An example diagram of the graph structure of the address reasoning logic is shown in FIG. Figure 2 The graph structure can infer the spatial address information associated with the task request information. By inferring the regional boundary data and regional satellite image data, and connecting the data, the connected data is filtered, and the address information spatial data is obtained from the knowledge graph, and then the inference result is output.
[0062] Furthermore, the target large model can utilize its own code generation capability to generate address reasoning code from the graph structure of the address reasoning logic, and execute the address reasoning code to obtain the first text address information associated with the task request information.
[0063] S104: perform information retrieval on the first text address information to determine a multimodal retrieval result of the first text address information, and determine first spatial description information of the first text address information based on the multimodal retrieval result.
[0064] In some implementations, in order to associate the spatial information of the address with the textual address information, information retrieval may be performed in a multimodal database based on the first textual address information to determine a multimodal retrieval result of the first textual address information.
[0065] Furthermore, spatial description information matching the first text address information can be determined from the multimodal search results as the first spatial description information of the first text address information, wherein the spatial description information includes but is not limited to: direction relationship, coordinates, longitude and latitude information of the text address, etc.
[0066] In some implementations, information retrieval can be performed based on the feature coding of the first text address information and the feature coding of the multimodal data, by calculating the similarity between the feature codings and selecting the feature coding with the highest similarity as the multimodal retrieval result.
[0067] In the information processing method provided by the embodiment of the present disclosure, the first prompt word of the task request information is obtained, and the first prompt word is reasoned and analyzed based on the pre-trained target large model to obtain the graph structure of the address reasoning logic. By converting the graph structure into an address reasoning code and executing the address reasoning code to obtain the first text address information associated with the task request information, the address information can be automatically reasoned, reducing the possibility of human error. Furthermore, the spatial information is integrated into the text address information, which improves the accuracy and clarity of the address information, so as to provide a more detailed description of the address and enhance the practicality of the address information.
[0068] Figure 3 This is an example flow chart of the information processing method provided by the embodiment of the present disclosure. Figure 3 As shown, the information processing method includes:
[0069] S301, obtaining task request information input by a user.
[0070] The relevant contents of step S301 can be found in the above embodiment and will not be described again here.
[0071] S302: Based on the task request information and according to the prompt word generation strategy, an initial second prompt word is obtained.
[0072] In some implementations, the initial second prompt word can be obtained from the task request information by predetermining a prompt word generation strategy. Optionally, the prompt word generation strategy can be a prompt template, and the task request information is input according to the prompt template to obtain the second prompt word.
[0073] By way of example, the generation strategy of prompt words may be: a detailed description of the task to be solved, a description of the way to present the information of the reasoning result, a suggested source of data, etc.
[0074] S303, based on the set reasoning mode of the target large model, the second prompt word is supplemented to obtain the first prompt word of the target large model.
[0075] Optionally, in order to enhance the reasoning ability of the model and improve the reasoning efficiency, the second prompt word may be modified according to the set reasoning mode to obtain the first prompt word with reasoning logic.
[0076] In some implementations, the node information of each reasoning step in the set reasoning mode is obtained, and the node information includes at least a first node and a second node. Optionally, the node information also includes: node type, data path, node description and other information.
[0077] It can be understood that each reasoning step has two forms, the first form is a file form, and the second form is an operation form, that is, each reasoning step can operate on the data and obtain the operation result of the data to store in the form of a file. The two forms of the reasoning step can be regarded as two nodes, namely the first node and the second node.
[0078] It can be understood that the processing result of the previous reasoning step is the input of the next reasoning step, and the first reasoning step is data acquisition, and the last reasoning step is to generate the result.
[0079] In some implementations, the second nodes should be connected through the first node to achieve the connection between the reasoning steps, that is, each reasoning step in the set reasoning mode can be connected based on the first node and the second node to obtain the spliced reasoning step.
[0080] For example, assume that the first node of reasoning step 1 is A1, and the second node is B1; the first node of reasoning step 2 is A2, and the second node is B2; the first node of reasoning step 3 is A3, and the second node is B3; based on the first node and the second node, the three reasoning steps are connected as A1-B1-A2-B2-A3-B3.
[0081] Further, a third prompt word is determined based on the concatenation reasoning step, and the third prompt word is concatenated with the second prompt word to obtain the first prompt word. In other words, the concatenation reasoning step can be used as the third prompt word, and the third prompt word is concatenated with the second prompt word to obtain the first prompt word.
[0082] S304, the target large model determines the graph structure of the address reasoning logic based on the first prompt word, generates address reasoning code based on the graph structure of the address reasoning logic, and executes the address reasoning code to obtain the first text address information associated with the task request information.
[0083] S305: perform information retrieval on the first text address information to determine a multimodal retrieval result of the first text address information, and determine first spatial description information of the first text address information based on the multimodal retrieval result.
[0084] The relevant contents of step S304 to step S305 can be found in the above embodiment and will not be repeated here.
[0085] In the information processing method provided by the embodiment of the present disclosure, by obtaining the initial second prompt information in the task request information, and determining the third prompt word according to the set reasoning mode, the second prompt word and the third prompt word are then spliced to obtain the first prompt word with reasoning logic, and the first prompt word is reasoned and analyzed based on the pre-trained target large model to obtain the graph structure of the address reasoning logic. By converting the graph structure into an address reasoning code and executing the address reasoning code to obtain the first text address information associated with the task request information, the address information can be automatically reasoned, reducing the possibility of human error. Furthermore, the spatial information is integrated into the text address information, which improves the accuracy and clarity of the address information, so as to describe the address in more detail and enhance the practicality of the address information.
[0086] Figure 4 This is an example flow chart of the information processing method provided by the embodiment of the present disclosure. Figure 4 As shown, the information processing method includes:
[0087] S401, obtaining task request information input by a user.
[0088] S402, based on the task request information, determine the first prompt word of the pre-trained target large model, and input the first prompt word into the target large model.
[0089] S403: The target large model determines the graph structure of the address reasoning logic based on the first prompt word.
[0090] The relevant contents of step S401 to step S403 can be found in the above embodiment and will not be repeated here.
[0091] S404, obtaining code constraints configured by the target large model, and based on the code constraints, performing code reasoning on the graph structure of the address reasoning logic to generate address reasoning code.
[0092] In some implementations, code constraints of the target large model can be set based on the configuration information of the target large model, such as the structure, function, input and output requirements, etc. Then, according to the code constraints, each step described in the graph structure of the reasoning logic is subjected to code reasoning using the code generation capability of the target large model itself, so as to obtain the address reasoning code corresponding to the graph structure of the address reasoning logic.
[0093] In some implementations, based on the code constraints configured in the target large model, code reasoning is performed on the graph structure of the address reasoning logic to obtain the address reasoning code, which can ensure that all constraints are strictly followed during the address reasoning process, thereby improving the accuracy and reliability of the reasoning results and further improving the efficiency of code generation.
[0094] For example, the code constraint condition may be:
[0095] (1) Maintain the names of given variables and paths;
[0096] (2) The returned result needs to be a complete Python code block;
[0097] (3) Data processing tools can be used to process spatial information data;
[0098] (4) Include comments and descriptions of input parameters and output return values;
[0099] (5) Do not load data repeatedly, load it in one function and use it globally;
[0100] (6) Try to use the latest Python packages and syntax;
[0101] (7) If multiple layers are involved, the layers should use the same projection rules;
[0102] (8) When table joins are involved, duplicate columns and rows need to be deleted.
[0103] In some implementations, the code generation capability of the large model can be fine-tuned and trained based on the reference reasoning code to obtain the target large model, and the code generation capability of the target large model can be used to generate address reasoning code with high accuracy and in compliance with code constraints.
[0104] S405: Execute the address inference code to obtain first text address information associated with the task request information.
[0105] S406: Perform information retrieval on the first text address information to determine a multimodal retrieval result of the first text address information, and determine first spatial description information of the first text address information based on the multimodal retrieval result.
[0106] The relevant contents of step S405 to step S406 can be found in the above embodiment and will not be repeated here.
[0107] In the information processing method provided by the embodiment of the present disclosure, the first prompt word of the task request information is obtained, and the first prompt word is reasoned and analyzed based on the pre-trained target large model to obtain the graph structure of the address reasoning logic. The graph structure is converted into the address reasoning code based on the code constraint condition, and the address reasoning code is executed to obtain the first text address information associated with the task request information, which can realize automatic reasoning of the address information and reduce the possibility of human error. Furthermore, the spatial information is integrated into the text address information, which improves the accuracy and clarity of the address information, so as to provide a more detailed description of the address and enhance the practicality of the address information.
[0108] Figure 5This is an example flow chart of the information processing method provided by the embodiment of the present disclosure. Figure 5 As shown, the information processing method includes:
[0109] S501: Obtain task request information input by a user.
[0110] The relevant contents of step S501 can be found in the above embodiment and will not be described again here.
[0111] S502: Perform intent recognition on the task request information to determine the task type.
[0112] In some implementations, in order to use different address identification methods, obtain address information associated with the task request information to improve the accuracy, efficiency, flexibility and adaptability of address identification. The task type can be determined first, and address identification is performed on the task request information according to the task type to determine the address information associated with the task request information.
[0113] Optionally, invalid information in the task request information is determined and filtered to obtain valid information in the task request information, and then the valid information is identified to determine whether the valid information carries address information.
[0114] Optionally, if the valid information does not carry address information, the task type is determined to be a reasoning task. For example, if the valid information filtered from the task request information is: opening a store near a school, with wide roads and good greenery, and the address information cannot be identified, the task type is determined to be a reasoning task.
[0115] S503, in response to the task type being a reasoning task, determining a first prompt word of a pre-trained target large model based on the task request information, and inputting the first prompt word into the target large model.
[0116] S504, the target large model determines the graph structure of the address reasoning logic based on the first prompt word, generates address reasoning code based on the graph structure of the address reasoning logic, and executes the address reasoning code to obtain the first text address information associated with the task request information.
[0117] S505: perform information retrieval on the first text address information to determine a multimodal retrieval result of the first text address information, and determine first spatial description information of the first text address information based on the multimodal retrieval result.
[0118] The relevant contents of step S503 to step S505 can be found in the above embodiment and will not be repeated here.
[0119] In the information processing method provided by the embodiment of the present disclosure, by determining that the task type is a reasoning task, the first prompt word of the task request information is obtained, and the first prompt word is reasoned and analyzed based on the pre-trained target large model to obtain the graph structure of the address reasoning logic. By converting the graph structure into an address reasoning code and executing the address reasoning code to obtain the first text address information associated with the task request information, the address information can be automatically reasoned, reducing the possibility of human error. Furthermore, the spatial information is integrated into the text address information, which improves the accuracy and clarity of the address information, so as to provide a more detailed description of the address and enhance the practicality of the address information.
[0120] like Figure 6 What is shown is a flow chart for address reasoning for reasoning tasks. By obtaining task request information, and determining the first prompt word based on the task request information, and then determining the graph structure of the address reasoning logic based on the first prompt word. Further, the address reasoning code is generated based on the graph structure using the code generation capability, and the address reasoning code is executed to obtain the first text address information associated with the task request information. In the process of reasoning the first text address information, the target large model is used to determine the graph structure of the address reasoning logic based on the first prompt word to generate the address reasoning code and execute the code to obtain the first text address information. After obtaining the first text address information, the first space description information corresponding to the first text address information can be obtained by performing information retrieval on the first text address information in the multimodal data and address space database.
[0121] Figure 7 This is an example flow chart of the information processing method provided by the embodiment of the present disclosure. Figure 7 As shown, the information processing method includes:
[0122] S701: If the valid information carries address information, determine that the task type is the first type.
[0123] For example, if the valid information filtered out from the task request information is: opening a store at XX location in XX county of this city, and the address information is determined to be XX county, XX location, then the task type of the task request information is the first type, that is, the first type represents the task type that carries address information.
[0124] S702, segment the valid information to obtain second text address information associated with the task request information.
[0125] In some implementations, in order to obtain accurate address information from valid information, the valid information can be segmented to automatically identify and extract the address information in the valid information, thereby improving the efficiency of address recognition and obtaining highly accurate second text address information.
[0126] In some implementations, the valid information is segmented by determining resource requirement information for information segmentation and selecting an appropriate segmentation method based on the resource requirement information. Alternatively, the segmentation method may be determined by setting a resource threshold and comparing the resource requirement information for information segmentation with the resource threshold.
[0127] Optionally, in response to the resource demand information being greater than a resource threshold, feature extraction is performed on the valid information to obtain feature data of the valid information, and conditional random field (CRF) segmentation is performed on the feature data to obtain second address information.
[0128] Optionally, in response to the resource requirement information being less than or equal to the resource threshold, a bidirectional encoder representation (Bidirectional Encoder Representations from Transformers, BERT) model based on a neural network is used to adjust the valid information to obtain adjusted valid information. Further, CRF segmentation is performed on the adjusted valid information to obtain the second address information.
[0129] Optionally, after obtaining the second address information, a compliance check may be performed on the second address information, and after the second address information passes the compliance check, the second address information may be standardized to obtain second address information in a standardized format.
[0130] S703: Perform information retrieval on the second text address information to determine second space description information of the second text address information.
[0131] In some implementations, after obtaining the first text address information and the second text address information, a multimodal information search can be performed on the text address information, which can integrate multiple information sources to improve the comprehensiveness of the search, and then retrieve the spatial description information of the text address information to realize intuitive display of the text address information.
[0132] Optionally, for any text address information in the first text address information and the second text address information, any text address information is encoded to obtain a first encoded representation, and then multimodal data stored in a database is obtained, and features are extracted from the multimodal data to obtain multiple feature vectors of the multimodal data, and the multiple feature vectors are fused to obtain a second encoded representation of the multimodal data.
[0133] Further, the similarity between the first coded representation and the second coded representation is obtained, and based on the similarity, a multimodal search result matching any text address information is determined from the database. Optionally, the similarities can be sorted, and the second coded representation with the greatest similarity to the first coded representation is selected as the multimodal search result.
[0134] In some implementations, after obtaining the spatial description information, an address space knowledge graph can be constructed based on the spatial description information to integrate the data, reduce data redundancy, improve data consistency and accuracy, and achieve standardization and normalization of address space data.
[0135] Optionally, an address space knowledge graph can be constructed based on the address elements and points of interest of the spatial description information, wherein the points of interest are used to describe the characteristic identifiers of the address, as well as the scope and related characteristics of the address information, and are considered as a supplement to the address elements.
[0136] Optionally, by obtaining address elements of the first spatial description information and the second spatial description information, and obtaining points of interest of the first spatial description information and the second spatial description information, and for any element in the address elements and the points of interest, determining the layers and connections between any elements.
[0137] Furthermore, any element is used as entity information to constitute a node in the address space knowledge graph, layers and connections are used as relationships between entities to constitute edges in the address space knowledge graph, and nodes and edges are connected to form the address space knowledge graph.
[0138] In the information processing method provided by the embodiment of the present disclosure, by determining that the task type of the task request information is the first type, segmenting the valid information in the task request information, and obtaining the second text address information, the address information can be automatically identified according to different task types, reducing the possibility of human error. Furthermore, the spatial information is integrated into the text address information, which improves the accuracy and clarity of the address information, so as to provide a more detailed description of the address and enhance the practicality of the address information.
[0139] like Figure 8 The figure shows a flow chart of address recognition for task request information. The task request information input by the user is obtained and the intent of the task request information is recognized to obtain valid information in the task request information, and the task type of the task request information is determined according to whether the valid information carries address information. If the valid information does not carry address information, the task type is determined to be a reasoning task; if the valid information carries address information, the task type is determined to be the first type.
[0140] When the task type is determined to be a reasoning task, the first prompt word of the pre-trained target large model is determined based on the task request information, and the graph structure of the address reasoning logic is determined. Then, based on the graph structure of the address reasoning logic, the address reasoning code is generated, and the target large model executes the address reasoning code to obtain the first text address information associated with the task request information.
[0141] When it is determined that the task type is the first type, the valid information is segmented to obtain the second text address information associated with the task request information.
[0142] Furthermore, for any text address information of the first text address information and the second text address information, information retrieval is performed on any text address information to determine the spatial description information of any text address information.
[0143] Corresponding to the information processing methods provided in the above-mentioned embodiments, an embodiment of the present disclosure further provides an information processing device. Since the information processing device provided in the embodiment of the present disclosure corresponds to the information processing methods provided in the above-mentioned embodiments, the implementation methods of the above-mentioned information processing methods are also applicable to the information processing device provided in the embodiment of the present disclosure and will not be described in detail in the following embodiments.
[0144] Fig. 9 This is a diagram showing an example of the structure of an information processing device provided by an embodiment of the present disclosure. Fig. 9 As shown, the information processing device includes: an acquisition module 901, a determination module 902, a reasoning module 903 and a retrieval module 904.
[0145] The acquisition module 901 is used to acquire the task request information input by the user;
[0146] A determination module 902, configured to determine a first prompt word of a pre-trained target large model based on the task request information, and input the first prompt word into the target large model;
[0147] The reasoning module 903 is used to determine the graph structure of the address reasoning logic based on the first prompt word by the target large model, generate address reasoning code based on the graph structure of the address reasoning logic, and execute the address reasoning code to obtain the first text address information associated with the task request information;
[0148] The retrieval module 904 is used to perform information retrieval on the first text address information, determine a multimodal retrieval result of the first text address information, and determine first spatial description information of the first text address information based on the multimodal retrieval result.
[0149] According to one embodiment of the present disclosure, the determination module 902 is also used to: based on the task request information, according to the prompt word generation strategy, obtain an initial second prompt word; based on the set reasoning mode of the target large model, supplement the second prompt word to obtain the first prompt word of the target large model.
[0150] According to one embodiment of the present disclosure, the determination module 902 is also used to: obtain node information of each reasoning step in the set reasoning mode, and the node information includes at least a first node and a second node; connect each reasoning step in the set reasoning mode based on the first node and the second node to obtain a spliced reasoning step; determine a third prompt word based on the spliced reasoning step, and splice the third prompt word and the second prompt word to obtain the first prompt word.
[0151] According to one embodiment of the present disclosure, the reasoning module 903 is further used to: obtain the code constraints of the target large model configuration, and based on the code constraints, perform code reasoning on the graph structure of the address reasoning logic to generate the address reasoning code.
[0152] According to one embodiment of the present disclosure, the determination module 902 is further used to: perform intent recognition on the task request information to determine the task type; in response to the task type being a reasoning task, determine the first prompt word of the pre-trained target large model based on the task request information.
[0153] According to one embodiment of the present disclosure, the determination module 902 is also used to: determine invalid information in the task request information, and filter the invalid information to obtain valid information in the task request information; identify the valid information to determine whether the valid information carries address information; if the valid information does not carry address information, determine that the task type is a reasoning task.
[0154] According to one embodiment of the present disclosure, the reasoning module 903 is also used to: if the valid information carries address information, determine that the task type is the first type; segment the valid information to obtain the second text address information associated with the task request information; perform information retrieval on the second text address information to determine the second spatial description information of the second text address information.
[0155] According to one embodiment of the present disclosure, the reasoning module 903 is also used to: determine resource requirement information for information segmentation; in response to the resource requirement information being greater than a resource threshold, perform feature extraction on the valid information to obtain feature data of the valid information; perform conditional random field (CRF) segmentation on the feature data to obtain the second address information.
[0156] According to one embodiment of the present disclosure, the device also includes: in response to the resource demand information being less than or equal to the resource threshold, using a neural network-based bidirectional encoder to represent the BERT model, adjusting the valid information to obtain adjusted valid information; performing CRF segmentation on the adjusted valid information to obtain the second address information.
[0157] According to one embodiment of the present disclosure, the retrieval module 904 is also used to: encode any text address information in the first text address information and the second text address information to obtain a first coded representation; obtain multimodal data stored in a database, perform feature extraction on the multimodal data to obtain multiple feature vectors of the multimodal data, and fuse the multiple feature vectors to obtain a second coded representation of the multimodal data; obtain the similarity between the first coded representation and the second coded representation, and determine a multimodal retrieval result matching any text address information from the database based on the similarity.
[0158] According to one embodiment of the present disclosure, the device also includes: obtaining address elements of the first spatial description information and the second spatial description information, and obtaining points of interest of the first spatial description information and the second spatial description information; for any element of the address elements and the points of interest, determining the layers and connections between any of the elements; using any of the elements as entity information to constitute a node in the address space knowledge graph, using the layers and connections as the relationship between entities to constitute an edge in the address space knowledge graph, and connecting the nodes and the edges to constitute an address space knowledge graph.
[0159] In the information processing device provided by the embodiment of the present disclosure, the first prompt word of the task request information is obtained, and the first prompt word is reasoned and analyzed based on the pre-trained target large model to obtain the graph structure of the address reasoning logic. By converting the graph structure into an address reasoning code and executing the address reasoning code to obtain the first text address information associated with the task request information, the address information can be automatically reasoned, reducing the possibility of human error. Furthermore, the spatial information is integrated into the text address information, which improves the accuracy and clarity of the address information, so as to provide a more detailed description of the address and enhance the practicality of the address information.
[0160] Fig.10 This is a diagram showing an example of the structure of an electronic device provided in an embodiment of the present disclosure. Fig.10 As shown, the electronic device 1000 may include: a transceiver 1001 , a processor 1002 , and a memory 1003 .
[0161] The processor 1002 executes the computer execution instructions stored in the memory, so that the processor 1002 executes the scheme in the above embodiment. The processor 1002 can be a general-purpose processor, including a central processing unit CPU, a network processor (network processor, NP), etc.; it can also be a digital signal processor DSP, an application-specific integrated circuit ASIC, a field programmable gate array FPGA or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components.
[0162] The memory 1003 is connected to the processor 1002 via a system bus and completes communication between them. The memory 1003 is used to store computer program instructions.
[0163] The transceiver 1001 may be used to obtain tasks to be executed and configuration information of the tasks to be executed.
[0164] The system bus can be a peripheral component interconnect (PCI) bus or an extended industry standard architecture (EISA) bus, etc. The system bus can be divided into an address bus, a data bus, a control bus, etc. For ease of representation, only one thick line is used in the figure, but it does not mean that there is only one bus or one type of bus. The transceiver is used to realize the communication between the database access device and other computers (such as clients, read-write libraries, and read-only libraries). The memory may include random access memory (RAM) and may also include non-volatile memory.
[0165] The electronic device provided in the embodiments of the present disclosure may be the terminal device of the above embodiments.
[0166] The disclosed embodiment also provides a chip for executing instructions, which is used to execute the technical solution of the information processing method in the above embodiment.
[0167] The embodiment of the present disclosure further provides a computer-readable storage medium, in which computer instructions are stored. When the computer instructions are executed on a computer, the computer executes the technical solution of the information processing method of the above embodiment.
[0168] The embodiments of the present disclosure also provide a computer program product, which includes a computer program stored in a computer-readable storage medium. At least one processor can read the computer program from the computer-readable storage medium. When the at least one processor executes the computer program, the technical solution of the information processing method in the above embodiments can be implemented.
[0169] Those skilled in the art will readily appreciate other embodiments of the present disclosure after considering the specification and practicing the invention disclosed herein. The present disclosure is intended to cover any variations, uses or adaptations of the present disclosure that follow the general principles of the present disclosure and include common knowledge or customary techniques in the art that are not disclosed in the present disclosure. The description and examples are to be considered exemplary only, and the true scope and spirit of the present disclosure are indicated by the following claims.
[0170] It should be understood that the present disclosure is not limited to the exact structures that have been described above and shown in the drawings, and that various modifications and changes may be made without departing from the scope thereof. The scope of the present disclosure is limited only by the appended claims.
Claims
1. An information processing method, characterized in that: The method comprises: Get the task request information entered by the user; Based on the task request information, determining a first prompt word of a pre-trained target large model, and inputting the first prompt word into the target large model; The target big model determines the graph structure of the address reasoning logic based on the first prompt word, generates address reasoning code based on the graph structure of the address reasoning logic, and executes the address reasoning code to obtain the first text address information associated with the task request information; An information search is performed on the first text address information to determine a multimodal search result of the first text address information, and first spatial description information of the first text address information is determined based on the multimodal search result.
2. The method according to claim 1, characterized in that The step of determining the first prompt word of the pre-trained target large model based on the task request information includes: Based on the task request information, and according to a prompt word generation strategy, an initial second prompt word is obtained; Based on the set reasoning mode of the target large model, the second prompt word is supplemented to obtain the first prompt word of the target large model.
3. The method according to claim 2, characterized in that The setting reasoning mode based on the target large model, supplementing the second prompt word to obtain the first prompt word of the large model, includes: Acquire node information of each reasoning step in the set reasoning mode, wherein the node information includes at least a first node and a second node; Connecting each reasoning step in the set reasoning mode based on the first node and the second node to obtain a concatenated reasoning step; A third prompt word is determined based on the concatenation reasoning step, and the third prompt word is concatenated with the second prompt word to obtain the first prompt word.
4. The method according to claim 1, characterized in that: The generating of address reasoning code based on the graph structure of the address reasoning logic includes: The code constraint conditions of the target large model configuration are obtained, and based on the code constraint conditions, code reasoning is performed on the graph structure of the address reasoning logic to generate the address reasoning code.
5. The method according to claim 1, characterized in that The step of determining the first prompt word of the pre-trained target large model based on the task request information further includes: Performing intent recognition on the task request information to determine the task type; In response to the task type being a reasoning task, a first prompt word of a pre-trained target large model is determined based on the task request information.
6. The method according to claim 5, characterized in that The performing intent recognition on the task request information to determine the task type includes: Determine invalid information in the task request information, and filter the invalid information to obtain valid information in the task request information; Identifying the valid information to determine whether the valid information carries address information; If the valid information does not carry address information, it is determined that the task type is a reasoning task.
7. The method according to claim 6, characterized in that The method further comprises: If the valid information carries address information, determining that the task type is the first type; Slicing the valid information to obtain second text address information associated with the task request information; Information retrieval is performed on the second text address information to determine second space description information of the second text address information.
8. The method according to claim 7, characterized in that The step of dividing the valid information to obtain the second address information includes: Determine the resource requirements for information segmentation; In response to the resource demand information being greater than a resource threshold, extracting features from the valid information to obtain feature data of the valid information; The feature data is segmented using a conditional random field (CRF) to obtain the second address information.
9. The method according to claim 8, characterized in that The method further comprises: In response to the resource requirement information being less than or equal to the resource threshold, using a bidirectional encoder representation BERT model based on a neural network to adjust the valid information to obtain adjusted valid information; The adjusted valid information is segmented by CRF to obtain the second address information.
10. The method according to claim 1 or 7, characterized in that: The information retrieval process includes: For any text address information of the first text address information and the second text address information, encoding the text address information to obtain a first coded representation; Acquire multimodal data stored in a database, perform feature extraction on the multimodal data to obtain a plurality of feature vectors of the multimodal data, and fuse the plurality of feature vectors to obtain a second coded representation of the multimodal data; The similarity between the first encoding representation and the second encoding representation is obtained, and based on the similarity, a multimodal retrieval result matching any one of the text address information is determined from the database.
11. The method according to claim 7, characterized in that The method further comprises: Acquire address elements of the first spatial description information and the second spatial description information, and acquire points of interest of the first spatial description information and the second spatial description information; For any element of the address element and the point of interest, determining the layer and the connection between the any element; Any of the elements is used as entity information to form a node in the address space knowledge graph, the layers and connections are used as relationships between entities to form an edge in the address space knowledge graph, and the nodes and edges are connected to form the address space knowledge graph.
12. An information processing device, characterized in that: The device comprises: An acquisition module is used to obtain task request information input by the user; A determination module, configured to determine a first prompt word of a pre-trained target large model based on the task request information, and input the first prompt word into the target large model; A reasoning module, configured to determine, by the target macromodel, a graph structure of an address reasoning logic based on the first prompt word, generate an address reasoning code based on the graph structure of the address reasoning logic, and execute the address reasoning code to obtain first text address information associated with the task request information; The retrieval module is used to perform information retrieval on the first text address information, determine a multimodal retrieval result of the first text address information, and determine first spatial description information of the first text address information based on the multimodal retrieval result.
13. The device according to claim 12, characterized in that The determining module is further used for: Based on the task request information, and according to a prompt word generation strategy, an initial second prompt word is obtained; Based on the set reasoning mode of the target large model, the second prompt word is supplemented to obtain the first prompt word of the target large model.
14. The device according to claim 13, characterized in that The determining module is further used for: Acquire node information of each reasoning step in the set reasoning mode, wherein the node information includes at least a first node and a second node; Connecting each reasoning step in the set reasoning mode based on the first node and the second node to obtain a concatenated reasoning step; A third prompt word is determined based on the concatenation reasoning step, and the third prompt word is concatenated with the second prompt word to obtain the first prompt word.
15. The device according to claim 12, characterized in that The reasoning module is further used for: The code constraint conditions of the target large model configuration are obtained, and based on the code constraint conditions, code reasoning is performed on the graph structure of the address reasoning logic to generate the address reasoning code.
16. The device according to claim 12, characterized in that The determining module is further used for: Performing intent recognition on the task request information to determine the task type; In response to the task type being a reasoning task, a first prompt word of a pre-trained target large model is determined based on the task request information.
17. The device according to claim 16, characterized in that The determining module is further used for: Determine invalid information in the task request information, and filter the invalid information to obtain valid information in the task request information; Identifying the valid information to determine whether the valid information carries address information; If the valid information does not carry address information, it is determined that the task type is a reasoning task.
18. The device according to claim 17, characterized in that The reasoning module is further used for: If the valid information carries address information, determining that the task type is the first type; Slicing the valid information to obtain second text address information associated with the task request information; Information retrieval is performed on the second text address information to determine second space description information of the second text address information.
19. The device according to claim 18, characterized in that The reasoning module is further used for: Determine the resource requirements for information segmentation; In response to the resource demand information being greater than a resource threshold, extracting features from the valid information to obtain feature data of the valid information; The feature data is segmented using a conditional random field (CRF) to obtain the second address information.
20. The device according to claim 19, characterized in that The device also includes: In response to the resource requirement information being less than or equal to the resource threshold, using a bidirectional encoder representation BERT model based on a neural network to adjust the valid information to obtain adjusted valid information; The adjusted valid information is segmented by CRF to obtain the second address information.
21. The device according to claim 12 or 18, characterized in that The retrieval module is further used for: For any text address information of the first text address information and the second text address information, encoding the text address information to obtain a first coded representation; Acquire multimodal data stored in a database, perform feature extraction on the multimodal data to obtain a plurality of feature vectors of the multimodal data, and fuse the plurality of feature vectors to obtain a second coded representation of the multimodal data; The similarity between the first encoding representation and the second encoding representation is obtained, and based on the similarity, a multimodal retrieval result matching any one of the text address information is determined from the database.
22. The device according to claim 18, characterized in that The device also includes: Acquire address elements of the first spatial description information and the second spatial description information, and acquire points of interest of the first spatial description information and the second spatial description information; For any element of the address element and the point of interest, determining the layer and the connection between the any element; Any of the elements is used as entity information to form a node in the address space knowledge graph, the layers and connections are used as relationships between entities to form an edge in the address space knowledge graph, and the nodes and edges are connected to form the address space knowledge graph.
23. An electronic device, characterized in that: include: A processor, and a memory communicatively connected to the processor; The memory stores computer-executable instructions; The processor executes the computer-executable instructions stored in the memory to implement the method according to any one of claims 1 to 11.
24. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer-executable instructions, which are used to implement the method according to any one of claims 1 to 11 when executed by a processor.
25. A computer program product, characterized in that The invention comprises a computer program, which implements the method according to any one of claims 1 to 11 when being executed by a processor.
Citation Information
Cited By
Address search method and device based on deep learning, storage medium and server
CN120277168A
