Geometric problem solving method, system and electronic device
By constructing a training database containing geometric calculations and proofs, and utilizing cross-modal neural networks to process geometric problems, the limitations of the solution range and poor applicability in existing technologies are solved, achieving efficient and accurate solutions to geometric problems.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- DMAI (GUANGZHOU) CO LTD
- Filing Date
- 2022-11-18
- Publication Date
- 2026-07-14
AI Technical Summary
In existing intelligent education products, the solution of geometry problems is divided into two main categories: calculation and proof. This results in a limited range of problems that can be solved, difficulty in writing rules, high costs, and poor applicability.
A training database containing geometric calculation problems and geometric proof problems is constructed. A cross-modal neural network is used to process the problem information, generate predicted solutions, and optimize the neural network through a loss function to achieve unified processing of geometric calculation and proof problems.
It achieves efficient and accurate solutions to geometric calculation and proof problems, avoids the need for additional models to determine the problem type, and improves the understanding and analytical capabilities of neural networks.
Smart Images
Figure CN116955419B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of intelligent education technology, specifically to a method, system, and electronic device for solving geometry problems. Background Technology
[0002] One application area of artificial intelligence is intelligent education. In previous intelligent education products, solving geometric problems was divided into two main categories: computation and proof. Related technologies treated these as two separate problems. For computational problems, some works proposed representing geometric problems using specific symbolic programs or logical forms. For geometric proofs, existing work mainly relies on carefully designed proof systems and forward chain search methods. Problem-solving algorithms for geometric problems typically use rule-based methods to parse the problem and perform some simple symbolic logic reasoning. This not only limits the scope of problem-solving for the product, but the cumbersome rule writing also increases the difficulty and cost of algorithm development, resulting in high implementation costs and poor applicability. Summary of the Invention
[0003] In view of this, the embodiments of this specification provide a method, system and electronic device for solving geometry problems, which can flexibly, efficiently and accurately handle calculation and proof-type geometry problems.
[0004] In a first aspect, embodiments of this specification provide a method for solving geometry problems, the method comprising:
[0005] Collect raw problem data and construct a training database based on the raw problem data, wherein the raw problem data includes geometric computation problem data and geometric proof problem data;
[0006] The question information is extracted from the training data items in the training database, and the question information is used as the input data of the cross-modal neural network. The cross-modal neural network is used to process the question information to generate predicted solution information for the question information.
[0007] By comparing the predicted solution information with the corresponding solution information of the question information in the training data item, the loss function of the cross-modal neural network is determined, and the cross-modal neural network is trained and optimized based on the loss function.
[0008] Optionally, a training database is constructed based on the original problem data, including:
[0009] The question portion of the original question data is used as the question information in the corresponding training data item;
[0010] The solution portion of the original problem data is converted into a solution process sequence, and the solution process sequence is used as the solution information in the corresponding training data item.
[0011] Optionally, the solution portion of the original problem data is converted into a solution process sequence, including:
[0012] Geometric entity elements, operator elements, and numeric elements are extracted from the solution data, and the solution process sequence is generated based on the geometric entity elements, operator elements, and numeric elements.
[0013] Optionally, after converting the solution portion of the geometric proof problem data into a solution process sequence, the method further includes:
[0014] The sequence of the solution process is adjusted according to the order of the proof reasons, operation symbols, and geometric elements.
[0015] Optionally, the question information is used as input data for a cross-modal neural network, and the cross-modal neural network is used to process the question information to generate predicted solution information, including:
[0016] Extract the text and chart data from the question information;
[0017] The text data and the chart data are encoded respectively to generate corresponding text feature vectors and chart feature vectors;
[0018] The text feature vector and the chart feature vector are concatenated and joint feature extraction is performed to generate a fused feature vector;
[0019] Multi-task joint decoding is performed on the fused feature vector to generate the predicted solution information.
[0020] Optionally, the text data is encoded, including:
[0021] A specific proportion of content is randomly selected from the text data for masking, and the masked text data is then encoded.
[0022] Optionally, after the cross-modal neural network training and optimization are completed, the method further includes:
[0023] Obtain the target geometry problem to be solved;
[0024] The trained and optimized cross-modal neural network is used to process the target geometry problem to generate predictive solution information for the target geometry problem;
[0025] The target geometry problem is solved based on the predicted solution information.
[0026] Optionally, in response to the target geometric problem being a geometric calculation type, the predicted solution information is a sequence of calculation processes;
[0027] Solving the target geometry problem based on the predicted solution information includes:
[0028] The calculation process is performed according to the sequence of calculations, and the calculation results are generated as the output.
[0029] In response to the target geometric problem being a geometric proof type, the predicted solution information is a sequence of proof processes;
[0030] Solving the target geometry problem based on the predicted solution information includes:
[0031] The proof process sequence is converted into readable proof process data as the output.
[0032] In a second aspect, embodiments of this specification also provide a geometry problem-solving system, the system comprising:
[0033] A database construction module is used to collect raw problem data and construct a training database based on the raw problem data, wherein the raw problem data includes geometric calculation problem data and geometric proof problem data;
[0034] The neural network processing module is used to extract question information from the training data items in the training database, use the question information as input data for the cross-modal neural network, and process the question information using the cross-modal neural network to generate predicted solution information based on the question information.
[0035] The training optimization module is used to determine the loss function of the cross-modal neural network by comparing the predicted solution information with the corresponding solution information of the question information in the training data item, and to train and optimize the cross-modal neural network based on the loss function.
[0036] The system is used to execute the solution method for the geometry problem.
[0037] In a third aspect, embodiments of this specification also provide an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the geometry problem-solving method as described in the first aspect.
[0038] As can be seen from the above, the geometry problem-solving method, system, and electronic device provided in the embodiments of this specification have the following beneficial technical effects:
[0039] A training database is constructed by collecting data on geometric calculation problems and geometric proof problems. A cross-modal neural network (CNN) is trained based on this database. During training, problem information from the training data items is extracted as input data for the CNN. The CNN performs encoding and decoding processing to generate predicted solutions. These predicted solutions are compared with the corresponding solutions from the training data. The CNN is then trained and optimized based on the comparison results. The training database contains both geometric calculation and geometric proof problem data. Training the CNN on this database enables unified processing of both types of problems without requiring an additional model to determine the problem type, thus avoiding the introduction of accumulated errors. The predicted solutions from the CNN are compared with the solutions from the training data items. Based on the comparison results, the CNN is trained and optimized to improve its ability to understand and analyze geometric problems, achieving efficient and accurate solutions for both geometric calculation and geometric proof problems. Attached Figure Description
[0040] The features and advantages of the invention will be more clearly understood by referring to the accompanying drawings, which are schematic and should not be construed as limiting the invention in any way. In the drawings:
[0041] Figure 1 This specification shows a schematic diagram of a method for solving geometry problems provided by one or more optional embodiments;
[0042] Figure 2 This diagram illustrates a method for building a training database in one or more optional embodiments of a geometry problem-solving method provided in this specification.
[0043] Figure 3 This diagram illustrates a method for processing problem information using the cross-modal neural network in one or more optional embodiments of a geometry problem-solving method provided in this specification;
[0044] Figure 4 This diagram illustrates a method for processing a geometric problem using a trained and optimized cross-modal neural network, as provided in one or more optional embodiments of this specification.
[0045] Figure 5 This specification shows a schematic diagram of the structure of a geometry problem-solving system provided by one or more optional embodiments;
[0046] Figure 6 This specification shows a schematic diagram of an electronic device structure for solving geometry problems, provided by one or more alternative embodiments. Detailed Implementation
[0047] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0048] One application area of artificial intelligence is intelligent education. In previous intelligent education products, solving geometric problems was divided into two main categories: computation and proof, treated as two separate problems in related technologies. For geometric computation problems, some works have proposed representing geometric problems using specific symbolic programs or logical forms; for geometric proof problems, existing work mainly relies on carefully designed proof systems and forward chain search methods. Algorithms for solving geometric problems typically use rule-based methods to parse the problem and perform some simple symbolic logical reasoning. Rule-based methods often require writing a large number of rules for specific problem types to ensure effective solutions, significantly increasing the difficulty and cost of algorithm development. Meanwhile, symbolic logical reasoning often fails to meet the requirements when dealing with highly complex logic, exhibiting poor applicability.
[0049] Based on the above problems, the purpose of the technical solution in this specification is to propose a method for solving geometry problems based on neural networks. This method involves constructing a training dataset that includes geometric calculation problem data and geometric proof problem data, and then training and optimizing a cross-modal neural network based on this dataset. This allows for the simultaneous processing of geometric calculation problems and geometric proof problems at both the data and model levels, thereby achieving efficient and accurate solutions to geometry problems.
[0050] Based on the above objectives, in one aspect, the embodiments of this specification provide a method for solving geometry problems.
[0051] like Figure 1 As shown, one or more optional embodiments of this specification provide a method for solving geometry problems, including:
[0052] S1: Collect raw problem data and construct a training database based on the raw problem data, wherein the raw problem data includes geometric calculation problem data and geometric proof problem data.
[0053] To simultaneously process geometric computation problems and geometric proof problems, data from both types of problems are collected as the original problem data to construct the training database.
[0054] When creating the training database, multiple training data items are generated based on multiple sets of original question data, and the training database is constructed based on these multiple training data items. Each set of original question data includes both question data and answer data.
[0055] Figure 2 This is a schematic diagram illustrating a method for constructing a training database based on the original problem data. (See diagram below.) Figure 2 As shown, in some specific embodiments, constructing a training database based on the original problem data can be achieved through the following steps:
[0056] S201: Use the question portion of the original question data as the question information in the corresponding training data item.
[0057] S202: Convert the solution portion of the original problem data into a solution process sequence, and use the solution process sequence as the solution information in the corresponding training data item.
[0058] Considering that the language used in the solution portion of the original problem data differs from that of general natural language, in order to improve the subsequent neural network model's ability to understand the solution information in the training data items, it is necessary to accurately extract the meaning of the solution information. In some optional embodiments, a mathematical expression-based approach is used to convert the solution portion into a sequence of solution processes as the solution information.
[0059] As a specific implementation method, the solution portion of the original problem data can be converted into a solution process sequence using the following scheme: extract geometric entity elements, operator elements, and number elements from the solution portion data, and organize and generate the solution process sequence based on the geometric entity elements, the operator elements, and the number elements.
[0060] In some alternative implementations, the extracted numeric elements can be replaced with data characters. For example, the numbers appearing in the solution data can be represented as NS. i , where i represents the order in which the numbers appear in the solution data. This method converts all the solution data in text form into mathematical expressions.
[0061] Considering that some geometric entity elements in the solution data contain common characters with specific geometric meanings, to ensure that the extracted geometric entity elements have accurate meaning expression, as a specific embodiment, character-level word segmentation is used to segment the solution data when extracting the geometric entity elements. Unlike word-by-word segmentation for natural text descriptions, the character-level word segmentation used for geometric entity elements in this embodiment can reflect the common character relationships between multiple geometric entity elements. For example, for the geometric entity elements OC and ∠OCA, directly using word-by-word segmentation results in OC and ∠OCA being treated as separate symbols with no connection between them. Using character-level word segmentation, three point elements O, C, and A are first divided, and then the geometric entity elements OC and ∠OCA are formed based on these point elements, sharing the common points O and C.
[0062] For the geometric calculation problem data, mathematical expressions are extracted from the solution portion of the data, and the resulting solution process sequence is a calculation process sequence. For the geometric proof problem data, mathematical expressions are extracted from the solution portion of the data, and the resulting solution process sequence is a proof process sequence.
[0063] It should be noted that, in addition to the geometric entity elements, the operator elements, and the numeric elements, the proof process sequence also includes a proof reason element. In some optional embodiments, the proof reason element in the proof process sequence can be replaced with a reason character. For example, the proof reason involved can be represented as R. j , where j represents the index of the proof reason in the predefined set of proof reasons.
[0064] Those skilled in the art will understand that the solution process for geometric computation problems generally follows a causal logical sequence. In contrast, the solution process for geometric proof problems requires deducing the result backward from known information, and its logical relationship differs from that of computation problems. Therefore, it is necessary to unify the proof process sequence for geometric proof problems with the computation process sequence for geometric computation problems. Thus, in some optional embodiments, after converting the solution portion of the geometric proof problem data into a solution process sequence, the solution process sequence is further adjusted according to the order of the proof cause, the operator, and the geometric elements.
[0065] S2: Extract question information from the training data items in the training database, use the question information as input data for the cross-modal neural network, and process the question information using the cross-modal neural network to generate predicted solution information for the question information;
[0066] The cross-modal neural network can be used to encode and decode the input data to output predictive problem-solving information corresponding to the input data.
[0067] Figure 3 This is a schematic diagram illustrating a method for processing the question information using the cross-modal neural network. For example... Figure 3 As shown, in some optional embodiments, the cross-modal neural network is used to process the question information, which can be implemented in the following ways:
[0068] S301: Extract the text data and chart data from the question information.
[0069] The question information includes text and charts, and the text and chart data can be extracted from the question information first.
[0070] S302: Encode the text data and the chart data respectively to generate corresponding text feature vectors and chart feature vectors.
[0071] The text data is represented as a text sequence. Treat each word in the text sequence as a word vector x i The text sequence can be encoded using the text encoder in the cross-modal neural network to obtain the corresponding hidden state vector as the text feature vector H. T .
[0072] For the chart data, the chart encoder in the cross-modal neural network is used to encode the chart data in order to extract and determine the corresponding chart feature vector H. D The graph encoder can apply the first four stages of convolutional layers in a classic residual neural network, namely the first four stages of ResNet-101, to process the graph data.
[0073] S303: Concatenate the text feature vector and the chart feature vector and perform joint feature extraction to generate a fused feature vector.
[0074] The chart feature vector H D With the text feature vector H T A preliminary vector representation of the entire geometric problem is obtained by concatenation and input into the cross-modal coding module in the cross-modal neural network for processing. The cross-modal coding module is then used to further extract high-dimensional features as the fused feature vector.
[0075] The cross-modal coding module consists of m Transformer layers, each containing a self-attention layer and a fully connected layer with residual connections. The number of Transformer layers can be set to m = 12. This cross-modal coding module enables further high-dimensional feature extraction from the text and chart feature vectors, achieving the extraction, fusion, and alignment of semantic information from the text and charts.
[0076] S304: Perform multi-task joint decoding on the fused feature vector to generate the predicted solution information.
[0077] The multi-task joint decoding module in the cross-modal neural network can be used to decode the fused feature vector to generate the predicted solution information.
[0078] The multi-task joint decoding module consists of another m-layer Transformer network structure. Compared to the cross-modal encoding module, the multi-task joint decoding module has an additional cross-attention layer. Through a self-attention mechanism, the multi-task joint decoding module iteratively focuses on the previously generated symbols and the fused feature vector output by the cross-modal encoding module, and then predicts the probability of future text symbols appearing. The multi-task joint decoding module can output a predicted sequence of problem-solving processes.
[0079] In some alternative embodiments, when encoding the text sequence using the text encoder in the cross-modal neural network, a specific proportion of content can be randomly selected from the text data for masking, and the masked text data can then be encoded.
[0080] For example, when training the cross-modal neural network, 30% of the geometry problem text in the training data can be randomly selected, masked, and input into the text encoder for processing. The cross-modal neural network then predicts the actual geometry problem content. It is understood that the specific percentage can be set to 30% and can be flexibly adjusted according to actual circumstances. Training in this way can enhance the cross-modal neural network's ability to understand and analyze geometry problems.
[0081] S3: By comparing the predicted solution information with the corresponding solution information of the question information in the training data item, the loss function of the cross-modal neural network is determined, and the cross-modal neural network is trained and optimized based on the loss function.
[0082] The loss function of the cross-modal neural network can be the negative log-likelihood estimation function:
[0083]
[0084] Where θ represents the parameters of the entire neural network model except for the graph encoder, since the parameters of the graph encoder are fixed; x represents the input data including the text data and the graph data.
[0085] In one or more optional embodiments of this specification, a method for solving geometry problems is provided, after the cross-modal neural network is trained and optimized, the trained and optimized cross-modal neural network is used to process the geometry problem to be solved.
[0086] Figure 4 This is a schematic diagram illustrating a method for processing a geometric problem using the trained and optimized cross-modal neural network. (See diagram for example.) Figure 4 As shown, the trained and optimized cross-modal neural network can be used to process the geometric problem to be solved in the following ways:
[0087] S401: Obtain the target geometry problem to be solved;
[0088] S402: The trained and optimized cross-modal neural network is used to process the target geometry problem to generate predictive solution information for the target geometry problem.
[0089] S403: Solve the target geometry problem based on the predicted solution information.
[0090] For a target geometric problem involving geometric calculations, the corresponding predicted solution information is a sequence of calculation processes. As a specific implementation, when solving the target geometric problem based on this sequence of calculation processes, operations can be performed according to the sequence to generate calculation results as output.
[0091] For geometry calculation problems, the operators based on the calculation process sequence are executed sequentially to obtain the calculated numerical result. If there are options, it is determined whether one of them is correct. Specifically, the N operation procedures {1,…, n After that, the program will perform calculations on it step by step. If g i The execution process will fail if there is a syntax error (e.g., the number of arguments does not match the current operator) or if the value executed does not match any of the options in the current problem. If all N programs fail, the executor will simply report "no result" without guessing an option.
[0092] For a target geometric problem of the geometric proof type, the corresponding predicted solution information is a sequence of proof processes. As a specific implementation, when solving the target geometric problem using the proof process sequence, the sequence is converted into readable proof process data as the output. Since the actual proof process may have multiple different answers, various proof sequences can also be generated.
[0093] As can be seen from the above, the proposed geometry problem-solving method collects data on geometric calculation problems and geometric proof problems to construct a training database. A cross-modal neural network is trained based on this database. During training, problem information from the training data items is extracted as input data for the cross-modal neural network. The network then performs encoding and decoding processing to generate predicted solutions. These predicted solutions are compared with the corresponding solutions from the training data. Based on the comparison results, the cross-modal neural network is trained and optimized. The training database contains both geometric calculation and geometric proof problem data. Training the cross-modal neural network on this database enables unified processing of both geometric calculation and geometric proof problems without requiring an additional model to determine the problem type, thus avoiding the introduction of accumulated errors. By comparing the predicted solutions from the cross-modal neural network with the solutions from the training data items, and optimizing the network based on the comparison results, the cross-modal neural network's ability to understand and analyze geometry problems is improved, enabling efficient and accurate solutions to both geometric calculation and geometric proof problems.
[0094] It should be noted that the methods of one or more embodiments of this specification can be executed by a single device, such as a computer or server. The methods of this embodiment can also be applied in a distributed scenario, where multiple devices cooperate to complete the task. In such a distributed scenario, one of these devices may execute only one or more steps of the methods of one or more embodiments of this specification, and the multiple devices will interact with each other to complete the method described.
[0095] It should be noted that the above description describes specific embodiments of this specification. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recorded in the claims may be performed in a different order than that shown in the embodiments and still achieve the desired results. Furthermore, the processes depicted in the drawings do not necessarily require the specific or sequential order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.
[0096] Based on the same inventive concept, and corresponding to any of the above embodiments, this specification also provides a geometry problem-solving system.
[0097] refer to Figure 5 The geometry problem-solving system includes:
[0098] A database construction module is used to collect raw problem data and construct a training database based on the raw problem data, wherein the raw problem data includes geometric calculation problem data and geometric proof problem data;
[0099] The neural network processing module is used to extract question information from the training data items in the training database, use the question information as input data for the cross-modal neural network, and process the question information using the cross-modal neural network to generate predicted solution information based on the question information.
[0100] The training optimization module is used to determine the loss function of the cross-modal neural network by comparing the predicted solution information with the corresponding solution information of the question information in the training data item, and to train and optimize the cross-modal neural network based on the loss function.
[0101] In a geometry problem-solving system provided in one or more optional embodiments of this specification, the database construction module is further configured to use the problem portion data in the original problem data as the problem information in the corresponding training data item; convert the solution portion data in the original problem data into a solution process sequence, and use the solution process sequence as the solution information in the corresponding training data item.
[0102] In a geometry problem-solving system provided in one or more optional embodiments of this specification, the database construction module is further configured to extract geometric entity elements, operator elements, and numeric elements from the solution portion data, and organize and generate the solution process sequence based on the geometric entity elements, the operator elements, and the numeric elements.
[0103] In a geometry problem-solving system provided in one or more optional embodiments of this specification, after the database construction module converts the solution portion data in the geometry proof problem data into a solution process sequence, it is further used to adjust the solution process sequence according to the order of proof reasons, operation symbols, and geometric elements.
[0104] In a geometry problem-solving system provided in one or more optional embodiments of this specification, the neural network processing module is further configured to extract text data and graph data from the problem information; encode the text data and graph data respectively to generate corresponding text feature vectors and graph feature vectors; concatenate the text feature vectors and graph feature vectors and perform joint feature extraction to generate a fused feature vector; and perform multi-task joint decoding on the fused feature vector to generate the predicted problem-solving information.
[0105] In a geometry problem-solving system provided by one or more optional embodiments of this specification, the neural network processing module is further configured to randomly select a specific proportion of content in the text data for masking processing, and then encode the masked text data.
[0106] The geometry problem-solving system provided in one or more optional embodiments of this specification further includes a geometry problem-solving module, which is used to obtain a target geometry problem to be solved; process the target geometry problem using the trained and optimized cross-modal neural network to generate predicted solution information for the target geometry problem; and solve the target geometry problem based on the predicted solution information.
[0107] In a geometry problem-solving system provided by one or more optional embodiments of this specification, for a target geometry problem of the geometry calculation type, the geometry problem-solving module is further configured to perform operations according to the corresponding calculation process sequence and generate calculation results as output results; for a target geometry problem of the geometry proof type, the geometry problem-solving module is further configured to convert the corresponding proof process sequence into readable proof process data as output results.
[0108] For ease of description, the above apparatus is described in terms of function, divided into various modules. Of course, when implementing one or more embodiments of this specification, the functions of each module can be implemented in one or more software and / or hardware.
[0109] The apparatus described above is used to implement the corresponding methods in the foregoing embodiments and has the beneficial effects of the corresponding method embodiments, which will not be repeated here.
[0110] Figure 6 This embodiment illustrates a more specific hardware structure of an electronic device, which may include a processor 1010, a memory 1020, an input / output interface 1030, a communication interface 1040, and a bus 1050. The processor 1010, memory 1020, input / output interface 1030, and communication interface 1040 are interconnected internally via the bus 1050.
[0111] The processor 1010 can be implemented using a general-purpose CPU (Central Processing Unit), microprocessor, application-specific integrated circuit (ASIC), or one or more integrated circuits, and is used to execute relevant programs to implement the technical solutions provided in the embodiments of this specification.
[0112] The memory 1020 can be implemented in the form of ROM (Read Only Memory), RAM (Random Access Memory), static storage device, dynamic storage device, etc. The memory 1020 can store the operating system and other applications. When the technical solutions provided in the embodiments of this specification are implemented by software or firmware, the relevant program code is stored in the memory 1020 and is called and executed by the processor 1010.
[0113] The input / output interface 1030 is used to connect input / output modules to realize information input and output. Input / output modules can be configured as components within the device (not shown in the figure) or externally connected to the device to provide corresponding functions. Input devices may include keyboards, mice, touchscreens, microphones, various sensors, etc., while output devices may include displays, speakers, vibrators, indicator lights, etc.
[0114] The communication interface 1040 is used to connect a communication module (not shown in the figure) to enable communication between this device and other devices. The communication module can communicate via wired means (such as USB, Ethernet cable, etc.) or wireless means (such as mobile network, WIFI, Bluetooth, etc.).
[0115] Bus 1050 includes a pathway for transmitting information between various components of the device, such as processor 1010, memory 1020, input / output interface 1030, and communication interface 1040.
[0116] It should be noted that although the above-described device only shows the processor 1010, memory 1020, input / output interface 1030, communication interface 1040, and bus 1050, in specific implementations, the device may also include other components necessary for normal operation. Furthermore, those skilled in the art will understand that the above-described device may only include the components necessary for implementing the embodiments of this specification, and not necessarily all the components shown in the figures.
[0117] The electronic devices described above are used to implement the corresponding methods in the foregoing embodiments and have the beneficial effects of the corresponding method embodiments, which will not be repeated here.
[0118] Based on the same inventive concept, corresponding to the methods of any of the above embodiments, this disclosure also provides a non-transitory computer-readable storage medium that stores computer instructions for causing the computer to execute the geometry problem-solving method as described in any of the above embodiments.
[0119] The computer-readable medium of this embodiment includes permanent and non-permanent, removable and non-removable media, and information storage can be implemented by any method or technology. Information can be computer-readable instructions, data structures, program modules, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic magnetic disk storage or other magnetic storage devices, or any other non-transfer medium that can be used to store information accessible by a computing device.
[0120] The computer instructions stored in the storage medium of the above embodiments are used to cause the computer to execute the geometry problem-solving method as described in any of the above embodiments, and have the beneficial effects of the corresponding method embodiments, which will not be repeated here.
[0121] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, it can include the processes of the embodiments of the methods described above. The storage medium can be a magnetic disk, optical disk, read-only memory (ROM), random access memory (RAM), flash memory, hard disk drive (HDD), or solid-state drive (SSD), etc.; the storage medium can also include combinations of the above types of memory.
[0122] The systems, devices, modules, or units described in the above embodiments can be implemented by computer chips or entities, or by products with certain functions. A typical implementation device is a computer. Specifically, a computer can be, for example, a personal computer, laptop computer, cellular phone, camera phone, smartphone, personal digital assistant, media player, navigation device, email device, game console, tablet computer, wearable device, or any combination of these devices.
[0123] For ease of description, the above devices are described separately by function as various units. Of course, in implementing this application, the functions of each unit can be implemented in one or more software and / or hardware.
[0124] Those skilled in the art will understand that embodiments of this specification can be provided as methods, systems, or computer program products. Therefore, this specification may take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this specification may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.
[0125] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.
[0126] This application can be described in the general context of computer-executable instructions, such as program modules, that are executed by a computer. Generally, program modules include routines, programs, objects, components, data structures, etc., that perform a specific task or implement a specific abstract data type. This application can also be practiced in distributed computing environments where tasks are performed by remote processing devices connected via a communication network. In distributed computing environments, program modules can reside in local and remote computer storage media, including storage devices.
[0127] The various embodiments in this specification are described in a progressive manner. Similar or identical parts between embodiments can be referred to interchangeably. Each embodiment focuses on describing the differences from other embodiments. In particular, the system embodiments are basically similar to the method embodiments, so the description is relatively simple; relevant parts can be referred to the descriptions in the method embodiments.
[0128] Those skilled in the art should understand that the discussion of any of the above embodiments is merely exemplary and is not intended to imply that the scope of this disclosure (including the claims) is limited to these examples; within the framework of this disclosure, the technical features of the above embodiments or different embodiments can also be combined, the steps can be implemented in any order, and there are many other variations of different aspects of one or more embodiments of this specification as described above, which are not provided in detail for the sake of brevity.
[0129] Although this disclosure has been described in conjunction with specific embodiments thereof, many substitutions, modifications, and variations of these embodiments will be apparent to those skilled in the art from the foregoing description. For example, other memory architectures (e.g., dynamic RAM (DRAM)) may be used with the embodiments discussed.
[0130] One or more embodiments of this specification are intended to cover all such substitutions, modifications, and variations that fall within the broad scope of the appended claims. Therefore, any omissions, modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of one or more embodiments of this specification should be included within the scope of protection of this disclosure.
Claims
1. A method for solving geometry problems, characterized in that, The method includes: Collect raw problem data and construct a training database based on the raw problem data, wherein the raw problem data includes geometric computation problem data and geometric proof problem data; The question information is extracted from the training data items in the training database, and the question information is used as the input data of the cross-modal neural network. The cross-modal neural network is used to process the question information to generate predicted solution information for the question information. By comparing the predicted problem-solving information with the corresponding solution information of the problem information in the training data item, the loss function of the cross-modal neural network is determined, and the cross-modal neural network is trained and optimized based on the loss function; A training database is constructed based on the original problem data, including: The question portion of the original question data is used as the question information in the corresponding training data item; The solution portion of the original problem data is converted into a solution process sequence, and the solution process sequence is used as the solution information in the corresponding training data item. After converting the solution portion of the geometric proof problem data into a solution process sequence, the method further includes: The sequence of the solution process is adjusted according to the order of the proof reasons, operation symbols, and geometric elements; After the cross-modal neural network training and optimization are completed, the following is also included: Obtain the target geometry problem to be solved; The trained and optimized cross-modal neural network is used to process the target geometry problem to generate predictive solution information for the target geometry problem; The target geometry problem is solved based on the predicted solution information; Since the target geometric problem is a geometric calculation type, the predicted solution information is a sequence of calculation processes; Solving the target geometry problem based on the predicted solution information includes: The calculation process is performed according to the sequence of calculations, and the calculation results are generated as the output. In response to the target geometric problem being a geometric proof type, the predicted solution information is a sequence of proof processes; Solving the target geometry problem based on the predicted solution information includes: The proof process sequence is converted into readable proof process data as the output. Convert the solution portion of the original problem data into a solution process sequence, including: Geometric entity elements, operator elements, and numeric elements are extracted from the solution data, and the solution process sequence is generated based on the geometric entity elements, operator elements, and numeric elements. When extracting the geometric entity elements, character-level word segmentation is used to segment the solution data.
2. The method according to claim 1, characterized in that, The question information is used as input data to a cross-modal neural network. The cross-modal neural network is then used to process the question information to generate predicted solution information, including: Extract the text and chart data from the question information; The text data and the chart data are encoded respectively to generate corresponding text feature vectors and chart feature vectors; The text feature vector and the chart feature vector are concatenated and joint feature extraction is performed to generate a fused feature vector; Multi-task joint decoding is performed on the fused feature vector to generate the predicted solution information.
3. The method according to claim 2, characterized in that, Encoding the text data includes: A specific proportion of content is randomly selected from the text data for masking, and the masked text data is then encoded.
4. A geometry problem-solving system, characterized in that, The system includes: A database construction module is used to collect raw problem data and construct a training database based on the raw problem data. The raw problem data includes geometric calculation problem data and geometric proof problem data. Constructing the training database based on the raw problem data includes: using the question portion data in the raw problem data as the question information in the corresponding training data item; converting the solution portion data in the raw problem data into a solution process sequence, and using the solution process sequence as the solution information in the corresponding training data item. It is also used to, after converting the solution portion of the geometric proof problem data into a solution process sequence, further include: adjusting the solution process sequence according to the order of the proof reason, the operation symbol, and the geometric element; The neural network processing module is used to extract question information from the training data items in the training database, use the question information as input data for the cross-modal neural network, and process the question information using the cross-modal neural network to generate predicted solution information based on the question information. The training optimization module is used to determine the loss function of the cross-modal neural network by comparing the predicted solution information with the corresponding solution information of the question information in the training data item, and to train and optimize the cross-modal neural network based on the loss function. After the cross-modal neural network training and optimization are completed, the following is also included: Obtain the target geometry problem to be solved; The trained and optimized cross-modal neural network is used to process the target geometry problem to generate predictive solution information for the target geometry problem; The target geometry problem is solved based on the predicted solution information; Since the target geometric problem is a geometric calculation type, the predicted solution information is a sequence of calculation processes; Solving the target geometry problem based on the predicted solution information includes: The calculation process is performed according to the sequence of calculations, and the calculation results are generated as the output. In response to the target geometric problem being a geometric proof type, the predicted solution information is a sequence of proof processes; Solving the target geometry problem based on the predicted solution information includes: The proof process sequence is converted into readable proof process data as the output. Convert the solution portion of the original problem data into a solution process sequence, including: Geometric entity elements, operator elements, and numeric elements are extracted from the solution data, and the solution process sequence is generated based on the geometric entity elements, operator elements, and numeric elements. When extracting the geometric entity elements, character-level word segmentation is used to segment the solution data.
5. An electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the method as described in any one of claims 1 to 3.