Information extraction method and device, equipment, storage medium and product
By determining the information extraction vector and text vector of the input text in the information extraction technology, and determining the target intention vector based on these vectors, the problem of low information extraction efficiency in the prior art is solved, and higher accuracy and efficiency are achieved.
Patent Information
- Application Number
- CN202411573408.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-11-05
- Publication Date
- 2025-02-18
AI Technical Summary
The existing information extraction technology has low accuracy in identifying key information in complex and unstructured texts and has insufficient understanding of user intentions, resulting in low information extraction efficiency.
By determining the information extraction vector and text vector of the input text, the target intention vector is determined based on these vectors, and finally the information extraction result is determined based on the target intention vector and text vector.
It improves the accuracy and efficiency of information extraction, and can better combine the intent information of the input text for information extraction.
Smart Images
Figure CN119476291A_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the field of natural language processing technology, and in particular to information extraction methods, devices, equipment, storage media and products. Background Art
[0002] With the continuous development of artificial intelligence and natural language processing technology, automated text information extraction technology has become a hot topic of research and application, especially in professional fields such as medicine and law. However, the complexity and diversity of text information in these fields pose challenges to the accuracy and efficiency of information extraction. To address this issue, related technologies use encoders to extract features from text information and determine information extraction results based on the feature extraction results. However, the accuracy of identifying key information in complex and unstructured texts needs to be improved, and the understanding of user intent is not accurate enough. Therefore, how to improve the efficiency of information extraction has become a technical problem that needs to be solved urgently. Summary of the invention
[0003] The main purpose of this application is to provide an information extraction method, device, equipment, storage medium and product, aiming to solve the technical problem of low efficiency of existing information extraction.
[0004] To achieve the above objectives, the present application proposes an information extraction method, which includes:
[0005] Determine the information extraction vector and text vector of the input text;
[0006] Determine a target intention vector according to the information extraction vector and the text vector;
[0007] An information extraction result corresponding to the input text is determined based on the target intention vector and the text vector.
[0008] Optionally, the step of determining the information extraction vector and the text vector of the input text includes:
[0009] Preprocess the input text to obtain the target text;
[0010] Extracting information from the target text to obtain an information extraction result;
[0011] Performing vector representation on the information extraction result through a pre-trained deep learning model to obtain an information extraction vector;
[0012] The target text is represented by a vector through a pre-trained deep learning model to obtain a text vector.
[0013] Optionally, the step of determining a target intention vector according to the information extraction vector and the text vector includes:
[0014] Determine an intention prediction vector according to the information extraction vector and the text vector;
[0015] Determine an intent candidate vector set according to the intent candidate set;
[0016] Determining an intent prediction loss based on the intent candidate vector set and the intent prediction vector;
[0017] A target intent vector is determined based on the intent prediction loss.
[0018] Optionally, the step of determining the intention prediction vector according to the information extraction vector and the text vector includes:
[0019] Splicing the information extraction vector and the text vector to obtain a splicing vector;
[0020] Performing dimension compression on the concatenated vector by using a preset information extraction model to obtain a compressed vector;
[0021] The compressed vector is encoded through a preset information extraction model to obtain an intention prediction vector.
[0022] Optionally, the step of determining the information extraction result corresponding to the input text based on the target intention vector and the text vector includes:
[0023] Determine a word segmentation vector based on a preset information extraction model, the target intent vector and the text vector;
[0024] The information extraction result corresponding to the input text is determined according to the word segmentation vector and the preset information extraction model.
[0025] Optionally, before the step of determining the word segmentation vector based on the preset information extraction model, the target intent vector and the text vector, the step further includes:
[0026] Determine the intention prediction loss of the sample data;
[0027] Determine the predicted entity labels and true labels of the sample data;
[0028] Determine a label loss value based on the predicted entity label and the true label;
[0029] An initial information extraction model is trained according to the intent prediction loss and the label loss value to obtain a preset information extraction model.
[0030] In addition, to achieve the above purpose, the present application also proposes an information extraction device, the information extraction device comprising:
[0031] A determination module, used for determining an information extraction vector and a text vector of an input text;
[0032] An intention vector determination module, used to determine a target intention vector according to the information extraction vector and the text vector;
[0033] An information extraction result determination module is used to determine the information extraction result corresponding to the input text based on the target intention vector and the text vector.
[0034] In addition, to achieve the above-mentioned purpose, the present application also proposes an information extraction device, which includes: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the computer program is configured to implement the steps of the information extraction method described above.
[0035] In addition, to achieve the above-mentioned purpose, the present application also proposes a storage medium, which is a computer-readable storage medium, and a computer program is stored on the storage medium. When the computer program is executed by a processor, the steps of the information extraction method described above are implemented.
[0036] In addition, to achieve the above-mentioned purpose, the present application also provides a computer program product, which includes a computer program, and when the computer program is executed by a processor, the steps of the information extraction method described above are implemented.
[0037] The present application determines the information extraction vector and text vector of the input text; determines the target intent vector based on the information extraction vector and the text vector; and determines the information extraction result corresponding to the input text based on the target intent vector and the text vector. Since the present application determines the information extraction result based on the text vector and the target intent vector corresponding to the input text, compared to the existing method of extracting features from the input text and determining the information extraction result based on the feature extraction result, the above method of the present application can extract information in combination with the intent information of the input text, thereby improving the accuracy of information extraction. BRIEF DESCRIPTION OF THE DRAWINGS
[0038] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present application and, together with the description, serve to explain the principles of the present application.
[0039] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, for ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.
[0040] Figure 1 A flowchart of the first embodiment of the information extraction method of this application is provided;
[0041] Figure 2 A flow chart of the second embodiment of the information extraction method of this application;
[0042] Figure 3 A schematic diagram of the model training process provided for Example 2 of the information extraction method of this application;
[0043] Figure 4 This is a schematic diagram of the module structure of the information extraction device according to an embodiment of the present application;
[0044] Figure 5 Schematic diagram of the device structure of the hardware operating environment involved in the information extraction method in the embodiment of the present application.
[0045] The purpose, features and advantages of this application will be further described in conjunction with the embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION
[0046] It should be understood that the specific embodiments described herein are only used to explain the technical solutions of the present application and are not used to limit the present application.
[0047] In order to better understand the technical solution of the present application, a detailed description will be given below in conjunction with the accompanying drawings and specific implementation methods.
[0048] The main solution of the embodiment of the present application is: determine the information extraction vector and text vector of the input text; determine the target intent vector based on the information extraction vector and the text vector; determine the information extraction result corresponding to the input text based on the target intent vector and the text vector. Since the present application determines the information extraction result based on the text vector and the target intent vector corresponding to the input text, compared with the existing method of extracting features from the input text and determining the information extraction result based on the feature extraction result, the above method of the present application can extract information in combination with the intent information of the input text, thereby improving the accuracy of information extraction.
[0049] It should be noted that the execution subject of this embodiment may be a computing service device with data processing, network communication and program running functions, such as a tablet computer, a personal computer, a mobile phone, etc., or an electronic device capable of realizing the above functions, an information extraction device, etc. The following takes the information extraction device as an example to illustrate this embodiment and the following embodiments.
[0050] Based on this, the present application embodiment provides an information extraction method, referring to Figure 1 , Figure 1 A flowchart of the first embodiment of the information extraction method of this application is provided.
[0051] In this embodiment, the information extraction method includes the following steps:
[0052] Step S10, determining the information extraction vector and text vector of the input text;
[0053] It should be noted that the input text may be a text that needs to be extracted. The information extraction vector of the input text may be determined by extracting information from the input text through a regularized matching technique, and then vectorizing the extraction result to obtain an information extraction vector. The text vector may be a text vector obtained by vectorizing the input text through a pre-trained deep learning model. The pre-trained deep learning model may be a pre-trained model for vectorizing text.
[0054] Furthermore, in order to improve the efficiency of information extraction, the step S10 may include: preprocessing the input text to obtain the target text;
[0055] Extracting information from the target text to obtain an information extraction result;
[0056] Performing vector representation on the information extraction result through a pre-trained deep learning model to obtain an information extraction vector;
[0057] The target text is represented by a vector through a pre-trained deep learning model to obtain a text vector.
[0058] It should be noted that the preprocessing of the input text to obtain the target text may be preprocessing operations such as word segmentation, stop word removal, part-of-speech tagging, etc. on the input text data to ensure that the obtained target text is suitable for subsequent model processing. The information extraction of the target text to obtain the information extraction result may be to use regularized matching technology to perform preliminary information extraction on the preprocessed target text to generate a candidate set of information extraction, that is, the information extraction result. The pre-trained deep learning model may be a pre-trained model for vectorizing data. The information extraction vector is a vector of n*256 dimensions, where n is the number of candidate information. The text vector is a vector of m*256 dimensions, where m is the number of words after word segmentation.
[0059] In a specific implementation, the information extraction device represents the information extraction result through a pre-trained deep learning model to generate an n*256-dimensional vector, where n is the number of candidate information.
[0060] V c =GTE(C)
[0061] Among them, C is the candidate set of information extraction, that is, the information extraction result, V c It is the vector representation of the information extraction result, which is an n*256-dimensional vector, where n is the number of candidate information. GTE refers to the pre-trained deep learning model used to generate text representation.
[0062] The training process of GTE is as follows: given a text x = {x1, x2, ..., x n}, where n is the number of tokens in the text. Use the pre-trained language model LM to obtain the deep context representation of each token:
[0063] h=LM(x)∈R n*d
[0064] Where d is the vector latitude.
[0065] The token representations are combined into a single text vector by operating on the first dimension through mean pooling:
[0066]
[0067] Set contrastive learning goals. Text representations are trained through contrastive learning, which aims to distinguish semantically related text pairs from unrelated text pairs. This method requires positive sample pairs and negative sample pairs. For a given question Que, the documents related to Que are marked as positive sample doc + , irrelevant documents are marked as negative samples doc - .
[0068] Set the loss function. In natural language processing, contrastive learning tasks usually use "information noise contrast estimation", which works by maximizing the similarity of positive sample pairs. Its original loss function is defined as follows:
[0069]
[0070] in is the jth document related to Que,
[0071]
[0072] s(x,y) represents the cosine similarity of the text embeddings of x and y.
[0073] This embodiment transforms the loss function to maximize the similarity of positive sample pairs while minimizing the similarity of negative sample pairs, and considers their weights, as follows:
[0074] L = -log(1 / (λ+(1-λ)Z)
[0075]
[0076] Among them, r is set to 0.01. is the jth document related to Que, is the i-th document that is irrelevant to Que, and λ is a weight value between 0 and 1. According to experiments, when λ is 0.6, the model converges fastest.
[0077] Training and evaluation: Model training consists of two stages: unsupervised contrastive pre-training and supervised contrastive fine-tuning. In the pre-training stage, a large number of text pairs are used for training; in the fine-tuning stage, high-quality text pairs with manual annotations are used for training.
[0078] The pre-trained language model LM is trained and optimized through the above steps to obtain a pre-trained deep learning model.
[0079] Step S20, determining a target intention vector according to the information extraction vector and the text vector;
[0080] It should be noted that the determination of the target intent vector based on the information extraction vector and the text vector can be to splice the information extraction vector and the text vector, compare the splicing result with each intent vector in the intent candidate set, obtain the cosine angle similarity between the intent vectors in the intent candidate set, and select the intent vector with the greatest similarity to the splicing result based on the cosine angle similarity, that is, the target intent vector. The intent candidate set is a set of intent categories pre-defined in the information extraction device, which represents the key information types that the user wants to extract from the text. These categories are obtained through domain knowledge analysis, expert consultation, data analysis and user feedback. During the information extraction process, the information extraction device uses these categories to generate an intent candidate set, and identifies the most relevant intent by calculating the similarity with the splicing result, thereby improving the accuracy and efficiency of information extraction.
[0081] Further, in order to improve the efficiency of information extraction, the step S20 may include: determining an intention prediction vector according to the information extraction vector and the text vector;
[0082] Determine an intent candidate vector set according to the intent candidate set;
[0083] Determining an intent prediction loss based on the intent candidate vector set and the intent prediction vector;
[0084] A target intent vector is determined based on the intent prediction loss.
[0085] It should be noted that the determination of the intention prediction vector according to the information extraction vector and the text vector may be performed by concatenating the information extraction vector and the text vector to form a (n+m)*256-dimensional vector S, and then forming a 1*256-dimensional vector S through a 1*(n+m)-dimensional fully connected layer of a preset information extraction model. *, and then processed by the encoder Encoder of the preset information extraction model to obtain the intent prediction vector I (1*256 dimensions). The determination of the intent candidate vector set based on the intent candidate set may be to vectorize each intent category in the intent candidate set to obtain the intent candidate vector set. Determining the intent prediction loss based on the intent candidate vector set and the intent prediction vector may be to calculate the cosine angle similarity between the intent prediction vector and each intention vector in the intent candidate vector set as the intent prediction loss. Determining the target intent vector based on the intent prediction loss may be to select the intent vector in the intent candidate vector set with the smallest intent prediction loss as the target intent vector. The preset information extraction model may be a pre-trained model for information extraction.
[0086] Step S30, determining the information extraction result corresponding to the input text based on the target intention vector and the text vector.
[0087] It should be noted that the information extraction result corresponding to the input text determined based on the target intention vector and the text vector can be obtained by splicing the target intention vector and the text vector, and predicting the information extraction result based on the spliced target intention vector and the text vector based on a pre-trained information extraction model.
[0088] Further, in order to improve the efficiency of information extraction, the step S30 may include: determining a word segmentation vector based on a preset information extraction model, the target intent vector and the text vector;
[0089] The information extraction result corresponding to the input text is determined according to the word segmentation vector and the preset information extraction model.
[0090] It should be noted that the word segmentation vector determined based on the preset information extraction model, the target intent vector and the text vector can be concatenated with the target intent vector and the text vector to obtain an m+1-dimensional vector, and the m+1-dimensional vector is processed by an encoder of the preset information extraction model to obtain a word segmentation vector. The information extraction result corresponding to the input text determined according to the word segmentation vector and the preset information extraction model can be predicted by the CRF layer of the preset information extraction model to predict the entity label of each word segmentation vector (token), that is, the information extraction result corresponding to the input text.
[0091] This embodiment determines the information extraction vector and text vector of the input text; determines the target intent vector based on the information extraction vector and the text vector; and determines the information extraction result corresponding to the input text based on the target intent vector and the text vector. Since this embodiment determines the information extraction result based on the text vector and the target intent vector corresponding to the input text, compared to the existing method of extracting features from the input text and determining the information extraction result based on the feature extraction result, the above method of this embodiment can extract information in combination with the intent information of the input text, thereby improving the accuracy of information extraction.
[0092] Based on the first embodiment of the present application, in the second embodiment of the present application, the same or similar contents as those in the above-mentioned embodiment 1 can be referred to the above introduction, and will not be repeated in the following. Figure 2 , Figure 2 The flowchart provided in the second embodiment of the information extraction method of the present application includes the following steps before step S10:
[0093] Step S001: Determine the intention prediction loss of sample data;
[0094] It should be noted that the determination of the intention prediction loss of the sample data can be determined by an initial information extraction model. Specifically, the sample information extraction vector (n*256 dimension) and the sample text vector (m*256 dimension) of the sample data are first determined. Specifically, the steps of determining the information extraction vector and the text vector of the input text can be referred to above. Then, the obtained sample information extraction vector and the sample text vector are concatenated to obtain a concatenated vector S. The concatenated vector S is input into the initial information extraction model. The fully connected layer of the initial information extraction model processes the concatenated vector S into a 1*256-dimensional vector S. * . Then, the intent prediction vector I (1*256 dimension) is obtained through the encoder of the initial information extraction model. Then, each intent in the intent candidate set is represented by a pre-trained deep learning model to form an intent candidate vector set, and the cosine angle similarity between the intent prediction vector I and the intent candidate vector set is calculated, that is, the intent prediction loss.
[0095] Step S002: Determine the predicted entity label and the true label of the sample data;
[0096] It should be noted that the determination of the predicted entity label and the true label of the sample data may be to determine the vector I in the intent candidate vector set that is most similar to the sample data according to the intent prediction loss. max , compare the sample text vector of the sample data with the vector I maxThe concatenation is then processed by the encoder of the initial information extraction model to obtain a token vector. The CRF layer of the initial information extraction model is then applied to predict the entity label of each token vector, that is, to obtain the predicted entity label. The real label can be a manually annotated label.
[0097] In the specific implementation, the information extraction device determines the sample information extraction vector (n*256 dimension) and sample text vector (m*256 dimension) of the sample data, and concatenates the n*256-dimensional sample information extraction vector with the m*256-dimensional sample text vector to form a (n+m)*256-dimensional vector S. Through a 1*(n+m)-dimensional fully connected layer, a 1*256-dimensional vector S* is formed. The intent prediction vector I is obtained by processing the encoder of the initial information extraction model. Each intent in the intent candidate set is represented by a pre-trained deep learning model to form an intent candidate vector set:
[0098] I candidates =GTE(IntentCandidates)
[0099] Calculate the intention prediction vector I and the intention candidate vector set I candidates The cosine angle similarity L between intent (i.e., intention to predict loss):
[0100] L intent =-log(CosineSimilarity(I,I candidates ))
[0101] Determine the intent with the greatest similarity:
[0102] I max =argmax(CosineSimilarity(I,I candidates ))
[0103] Among them, IntentCandidates is the intent candidate set, I candidates is the set of intent candidate vectors, I is the intent prediction vector, and I max is the most similar intent vector, L intent is the intention prediction loss (cosine angle similarity), CosineSimilarity represents the cosine similarity function, and GTE() represents the pre-trained deep learning model.
[0104] Compare the sample text vector of the sample data with the vector I max Concatenate to form an m+1 dimensional vector:
[0105] S enhanced =Concatenate(Vt , I max )
[0106] Among them, V t The sample text vector representing the sample data, Concatenate represents concatenation, S enbanced Represents the concatenated enhanced vector.
[0107] The token vector is obtained by processing it through the encoder of the initial information extraction model, and then the CRF layer of the initial information extraction model is applied to predict the entity label of each token.
[0108] V token =Encoder(S enbanced )
[0109] Among them, V token Represents a token vector.
[0110] Step S003: determining a label loss value based on the predicted entity label and the true label;
[0111] It should be noted that the determining of the label loss value based on the predicted entity label and the true label may be to use the negative log-likelihood loss function of CRF to measure the difference between the predicted label sequence and the true label sequence:
[0112] L entity =CRFLoss(V token , Y true )
[0113] Among them, L entity is the entity prediction loss, i.e., the label loss value, Y true Represents the true entity label of each token. CRFLoss represents the negative log-likelihood loss function of CRF.
[0114] Step S004: training an initial information extraction model according to the intent prediction loss and the label loss value to obtain a preset information extraction model.
[0115] It should be noted that the initial information extraction model is trained according to the intention prediction loss and the label loss value to obtain the preset information extraction model, which can be performed by cyclically executing the above steps of this embodiment starting from step S001, and updating the parameters of the initial information extraction model through back propagation and optimization algorithm until the intention prediction loss L intent and entity prediction loss L entity The absolute value changes are all less than 10e-10, achieving model convergence and obtaining the preset information extraction model. Figure 3 , Figure 3A schematic diagram of the model training process provided for Example 2 of the information extraction method of this application. This example allows the model to predict sentence categories and information extraction in a mutual manner. This mechanism enhances the model's in-depth understanding of text content and improves the ability to extract key information. It can achieve information complementarity and performance improvement between tasks.
[0116] This embodiment determines the intention prediction loss of sample data; determines the predicted entity label and the true label of the sample data; determines the label loss value based on the predicted entity label and the true label; trains the initial information extraction model according to the intention prediction loss and the label loss value to obtain a preset information extraction model. In the process of cyclic training and loss minimization, this embodiment adopts the technical solution of sentence classification, and its role is reflected in the following aspects: Enhance context understanding: Through sentence classification, the model can have a deeper understanding of the context and intention of the text, which is crucial for accurately extracting key information; Optimize information extraction: The results of sentence classification serve as auxiliary information for information extraction tasks, helping the model to more accurately identify and extract key information; Improve the generalization ability of the model: Consider sentence classification and information extraction at the same time during the training process, so that the model can learn richer feature representations, thereby showing better generalization ability on different texts and tasks. When the model meets the preset stop condition, the training process ends, and the model parameters at this time are determined to be optimal. Finally, the model can automatically classify the input text, extract key information from it, and output it as structured data for further analysis and application.
[0117] It should be noted that the above examples are only used to understand the present application and do not constitute a limitation on the information extraction method of the present application. More simple transformations based on this technical concept are all within the scope of protection of the present application.
[0118] This application also provides an information extraction device, please refer to Figure 4 , the information extraction device comprises:
[0119] A determination module 10, used to determine an information extraction vector and a text vector of an input text;
[0120] An intention vector determination module 20, configured to determine a target intention vector according to the information extraction vector and the text vector;
[0121] The information extraction result determination module 30 is used to determine the information extraction result corresponding to the input text based on the target intention vector and the text vector.
[0122] This embodiment determines the information extraction vector and text vector of the input text; determines the target intent vector based on the information extraction vector and the text vector; and determines the information extraction result corresponding to the input text based on the target intent vector and the text vector. Since this embodiment determines the information extraction result based on the text vector and the target intent vector corresponding to the input text, compared to the existing method of extracting features from the input text and determining the information extraction result based on the feature extraction result, the above method of this embodiment can extract information in combination with the intent information of the input text, thereby improving the accuracy of information extraction.
[0123] The information extraction device provided by the present application adopts the information extraction method in the above embodiment, which can solve the technical problem of low efficiency of existing information extraction. Compared with the prior art, the beneficial effects of the information extraction device provided by the present application are the same as the beneficial effects of the information extraction method provided by the above embodiment, and other technical features in the information extraction device are the same as the features disclosed in the above embodiment method, which will not be repeated here.
[0124] The present application provides an information extraction device, which includes: 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 so that the at least one processor can execute the information extraction method in the above-mentioned embodiment one.
[0125] Reference below Figure 5 , which shows a schematic diagram of the structure of an information extraction device suitable for implementing the embodiment of the present application. The information extraction device in the embodiment of the present application may include but is not limited to mobile terminals such as mobile phones, laptop computers, digital broadcast receivers, PDAs (Personal Digital Assistants), PADs (Portable Application Descriptions), PMPs (Portable Media Players), vehicle-mounted terminals (such as vehicle-mounted navigation terminals), etc., and fixed terminals such as digital TVs, desktop computers, etc. Figure 5 The information extraction device shown is merely an example and should not bring any limitation to the functions and scope of use of the embodiments of the present application.
[0126] like Figure 5As shown, the information extraction device may include a processing device 1001 (e.g., a central processing unit, a graphics processing unit, etc.), which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM: Read Only Memory) 1002 or a program loaded from a storage device 1003 to a random access memory (RAM: Random Access Memory) 1004. Various programs and data required for the operation of the information extraction device are also stored in the RAM 1004. The processing device 1001, the ROM 1002, and the RAM 1004 are connected to each other via a bus 1005. An input / output (I / O) interface 1006 is also connected to the bus. Typically, the following systems can be connected to the I / O interface 1006: input devices 1007 including, for example, a touch screen, a touchpad, a keyboard, a mouse, an image sensor, a microphone, an accelerometer, a gyroscope, etc.; output devices 1008 including, for example, a liquid crystal display (LCD), a speaker, a vibrator, etc.; storage devices 1003 including, for example, a magnetic tape, a hard disk, etc.; and communication devices 1009. The communication device 1009 can allow the information extraction device to communicate wirelessly or wired with other devices to exchange data. Although the figure shows an information extraction device with various systems, it should be understood that it is not required to implement or have all the systems shown. More or fewer systems may be implemented or have alternatively.
[0127] In particular, according to the embodiments disclosed in the present application, the process described above with reference to the flowchart can be implemented as a computer software program. For example, the embodiments disclosed in the present application include a computer program product, which includes a computer program carried on a computer-readable medium, and the computer program includes a program code for executing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from a network through a communication device, or installed from a storage device 1003, or installed from a ROM 1002. When the computer program is executed by the processing device 1001, the above-mentioned functions defined in the method of the embodiment disclosed in the present application are executed.
[0128] The information extraction device provided by the present application adopts the information extraction method in the above embodiment, which can solve the technical problem of low efficiency of existing information extraction. Compared with the prior art, the beneficial effects of the information extraction device provided by the present application are the same as the beneficial effects of the information extraction method provided by the above embodiment, and other technical features in the information extraction device are the same as the features disclosed in the method of the previous embodiment, which will not be repeated here.
[0129] It should be understood that the various parts disclosed in this application can be implemented by hardware, software, firmware or a combination thereof. In the description of the above embodiments, specific features, structures, materials or characteristics can be combined in any one or more embodiments or examples in a suitable manner.
[0130] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art who is familiar with the present technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.
[0131] The present application provides a computer-readable storage medium having computer-readable program instructions (ie, computer programs) stored thereon, wherein the computer-readable program instructions are used to execute the information extraction method in the above-mentioned embodiment.
[0132] The computer-readable storage medium provided in the present application may be, for example, a USB flash drive, but is not limited to electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, systems or devices, or any combination of the above. More specific examples of computer-readable storage media may include, but are not limited to: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In this embodiment, the computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in combination with an instruction execution system, system or device. The program code contained on the computer-readable storage medium may be transmitted using any appropriate medium, including but not limited to: wires, optical cables, RF (Radio Frequency), etc., or any suitable combination of the above.
[0133] The computer-readable storage medium may be included in the information extraction device; or may exist independently without being assembled into the information extraction device.
[0134] Computer program code for performing the operations of the present application may be written in one or more programming languages or a combination thereof, including object-oriented programming languages such as Java, Smalltalk, C++, and conventional procedural programming languages such as "C" or similar programming languages. The program code may be executed entirely on the user's computer, partially on the user's computer, as a separate software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In the case of a remote computer, the remote computer may be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., via the Internet using an Internet service provider).
[0135] The flow chart and block diagram in the accompanying drawings illustrate the possible architecture, function and operation of the system, method and computer program product according to various embodiments of the present application. In this regard, each square box in the flow chart or block diagram can represent a module, a program segment or a part of a code, and the module, the program segment or a part of the code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the square box can also occur in a sequence different from that marked in the accompanying drawings. For example, two square boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each square box in the block diagram and / or flow chart, and the combination of the square boxes in the block diagram and / or flow chart can be implemented with a dedicated hardware-based system that performs a specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.
[0136] The modules involved in the embodiments described in this application may be implemented by software or hardware, wherein the name of the module does not constitute a limitation on the unit itself in some cases.
[0137] The readable storage medium provided in the present application is a computer-readable storage medium, which stores computer-readable program instructions (i.e., computer programs) for executing the above-mentioned information extraction method, and can solve the technical problem of low efficiency of existing information extraction. Compared with the prior art, the beneficial effects of the computer-readable storage medium provided in the present application are the same as the beneficial effects of the information extraction method provided in the above-mentioned embodiment, and will not be described in detail here.
[0138] The present application also provides a computer program product, including a computer program, which implements the steps of the above-mentioned information extraction method when executed by a processor.
[0139] The computer program product provided by this application can solve the technical problem of low efficiency of existing information extraction. Compared with the prior art, the beneficial effects of the computer program product provided by this application are the same as the beneficial effects of the information extraction method provided by the above embodiment, which will not be repeated here.
[0140] The above descriptions are only some embodiments of the present application, and are not intended to limit the patent scope of the present application. All equivalent structural changes made using the contents of the present application specification and drawings under the technical concept of the present application, or direct / indirect applications in other related technical fields are included in the patent protection scope of the present application.
Claims
1. An information extraction method, characterized in that: The information extraction method comprises the following steps: Determine the information extraction vector and text vector of the input text; Determine a target intention vector according to the information extraction vector and the text vector; An information extraction result corresponding to the input text is determined based on the target intention vector and the text vector.
2. The information extraction method according to claim 1, characterized in that: The step of determining the information extraction vector and the text vector of the input text comprises: Preprocess the input text to obtain the target text; Extracting information from the target text to obtain an information extraction result; Performing vector representation on the information extraction result through a pre-trained deep learning model to obtain an information extraction vector; The target text is represented by a vector through a pre-trained deep learning model to obtain a text vector.
3. The information extraction method according to claim 1, characterized in that: The step of determining the target intention vector according to the information extraction vector and the text vector comprises: Determine an intention prediction vector according to the information extraction vector and the text vector; Determine an intent candidate vector set according to the intent candidate set; Determining an intent prediction loss based on the intent candidate vector set and the intent prediction vector; A target intent vector is determined based on the intent prediction loss.
4. The information extraction method according to claim 3, characterized in that: The step of determining the intention prediction vector according to the information extraction vector and the text vector comprises: Splicing the information extraction vector and the text vector to obtain a splicing vector; Performing dimension compression on the concatenated vector by using a preset information extraction model to obtain a compressed vector; The compressed vector is encoded through a preset information extraction model to obtain an intention prediction vector.
5. The information extraction method according to any one of claims 1 to 4, characterized in that: The step of determining the information extraction result corresponding to the input text based on the target intention vector and the text vector comprises: Determine a word segmentation vector based on a preset information extraction model, the target intent vector and the text vector; The information extraction result corresponding to the input text is determined according to the word segmentation vector and the preset information extraction model.
6. The information extraction method according to claim 5, characterized in that: Before the step of determining the word segmentation vector based on the preset information extraction model, the target intent vector and the text vector, the step further includes: Determine the intention prediction loss of the sample data; Determine the predicted entity labels and true labels of the sample data; Determine a label loss value based on the predicted entity label and the true label; An initial information extraction model is trained according to the intent prediction loss and the label loss value to obtain a preset information extraction model.
7. An information extraction device, characterized in that: The information extraction device comprises: A determination module, used for determining an information extraction vector and a text vector of an input text; An intention vector determination module, used to determine a target intention vector according to the information extraction vector and the text vector; An information extraction result determination module is used to determine the information extraction result corresponding to the input text based on the target intention vector and the text vector.
8. An information extraction device, characterized in that: The device comprises: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the computer program is configured to implement the steps of the information extraction method according to any one of claims 1 to 6.
9. A storage medium, characterized in that: The storage medium is a computer-readable storage medium, and a computer program is stored on the storage medium. When the computer program is executed by a processor, the steps of the information extraction method according to any one of claims 1 to 6 are implemented.
10. A computer program product, characterized in that The computer program product comprises a computer program, and when the computer program is executed by a processor, the steps of the information extraction method according to any one of claims 1 to 6 are implemented.