Quick anti-deductive learning method and device for structured text recognition

Abstract

The invention discloses a quick anti-deductive learning method and device for structured text recognition, and the method comprises the steps: learning an obtained text symbol sequence in a knowledge base through a machine learning model, selecting a text symbol with the highest probability as probability prediction information, and predicting the relation between symbols in the text symbol sequence through a neural network model. And obtaining a probability sequence belonging to a constant by combining the relation between the probability prediction information and the symbols, generating a discrete Boolean sequence based on the probability sequence belonging to the constant, and inputting the discrete Boolean sequence into the original anti-crimson learning architecture for logical reasoning. According to the method, the anti-deductive learning is endowed with text symbol-based meta-reasoning capability similar to that of human experts, and considerable results can be obtained only through a small number of trial and error attempts. In addition, the quick anti-deductive learning for structured text recognition can be more efficient and rigorous, the number of times of operation of a logical reasoning module in the anti-deductive learning is effectively reduced, and the efficiency of the anti-deductive learning is remarkably improved.

Description

Technical Field

[0001] The present invention relates to a fast abductive learning method and device for structured text recognition, belonging to the technical field of model data processing in artificial intelligence technology. Background Art

[0002] Abductive learning refers to starting from an incomplete observation to obtain the most likely explanation for a particular set of interest. When previous abductive learning techniques are applied to process structured text sequences, in the worst case, they are often accompanied by high computational costs. Therefore, it is particularly crucial to achieve high-efficiency computing in the interaction process between the two fields of machine learning and logical reasoning. Within the framework of abductive learning, the logical reasoning module is responsible for classifying the symbols predicted by the machine learning model into unreliable parts (regarded as variables) and reliable parts (regarded as constants), and then stringing together the rules in the knowledge base based on this, and further inferring the symbols regarded as variables using the constants, and correcting these symbols regarded as variables and then feeding them back to the machine learning model for updating. This process can essentially be regarded as an optimization problem, where the optimization variables are Boolean variables, which determine which predicted symbols should be used as variables in the reasoning and which should be used as constants, and the optimization goal is to maximize the number of final correct logical reasoning results.

[0003] In view of the fact that the optimization problem of processing structured text sequences involves the processing of both numerical values and symbols at the same time, but there is currently a lack of efficient optimization techniques. It should be noted that during the entire optimization process, the execution times of machine learning and logical reasoning do not correspond one-to-one, but rather show a one-to-many relationship, which will result in logical reasoning occupying most of the time. Usually, hundreds of logical reasoning operations are required to correspond to one machine learning process. Therefore, significantly reducing the number of attempts of Boolean variables in logical reasoning is crucial for greatly improving the efficiency of abductive learning.

[0004] However, for this optimization problem, traditional abductive learning methods adopt gradient-free optimization methods, such as the search-based algorithm POSS and the sample-based algorithm RACOS, etc. Although these methods are different from the way human experts combine induction and reasoning, they are often accompanied by a large number of ineffective operations, resulting in a waste of resources. Specifically, compared with human experts, the reasons for the low efficiency of existing algorithms are mainly attributed to three aspects: failure to make full use of prediction information starting from a random state; failure to effectively utilize the relationships between symbols; and failure to accumulate experience in successful abductive learning processes. Therefore, there is an urgent need to provide a fast abductive learning method for structured text recognition. Summary of the Invention

[0005] The content part of this application is used to introduce concepts in a brief form, which will be described in detail in the specific implementation part later. The content part of this application is not intended to identify the key features or essential features of the claimed technical solution, nor is it intended to limit the scope of the claimed technical solution.

[0006] Aiming at the problems and deficiencies in the prior art, the purpose of the present invention is to provide a fast abductive learning method and device for structured text recognition. By using the probability prediction results of a machine learning model as input, and combining the relationships between symbol sequences, each probability to be corrected in the input is output as a variable. Secondly, a method of generating a discrete Boolean sequence from a continuous probability sequence is adopted to determine specific symbols in the sequence as variables in logical reasoning and perform the abductive process. This effectively reduces the number of runs of the logical reasoning module in abductive learning and significantly improves the efficiency of abductive learning. It is used to solve the problems raised in the above background technology.

[0007] To achieve the above purpose, the present invention provides the following technical solutions:

[0008] The present invention discloses a fast abductive learning method for structured text recognition, including the following steps:

[0009] Step 1, obtain the original abductive learning architecture and knowledge base to be adjusted;

[0010] Step 2, use a machine learning model to predict the probabilities of text symbol sequences in the knowledge base to obtain probability prediction information of text symbols;

[0011] Step 3, use a neural network model to predict the relationships between symbols in the text symbol sequence;

[0012] Step 4, then use the machine learning model to combine the probability prediction information and the relationships between symbols to predict a probability sequence belonging to constants;

[0013] Step 5, generate a discrete Boolean sequence based on the probability sequence belonging to constants and input it into the original abductive learning architecture for logical reasoning.

[0014] Preferably, step 2 further includes the following steps:

[0015] Step 2.1, add the machine learning model to the original abductive learning architecture;

[0016] Step 2.2, use the machine learning model to predict the probabilities of text symbol sequences in the knowledge base;

[0017] Step 2.3, retain the probability prediction information of each text symbol to determine its reliability;

[0018] Step 2.4, obtain the category of the text symbol with the highest probability as the final text symbol category.

[0019] Preferably, in step 3, after performing logical reasoning in step 5, use the Boolean variable with a higher consistency between the knowledge base and the discrete Boolean sequence as the supervision information, and input the supervision information into the machine learning model for updating until convergence.

[0020] Preferably, step 5 specifically includes the following steps:

[0021] Step 5.1, generate an initial discrete Boolean sequence according to the probability sequence assignment of the constants;

[0022] Step 5.2, perform a heap search based on the initial discrete Boolean sequence to obtain a subsequent discrete Boolean sequence;

[0023] Step 5.3, further sort the subsequent discrete Boolean sequences according to the probability;

[0024] Step 5.4, each time select the subsequent discrete Boolean sequence with the highest probability as the discrete Boolean sequence output.

[0025] Preferably, in step 5.1, generate an initial discrete Boolean sequence according to the probability sequence assignment of the constants,

[0026] and its assignment rule is to assign a probability greater than 0.5 in the probability sequence of the constants as 1, otherwise assign 0.

[0027] Preferably, in step 2, predict the probability of the text symbol sequence in the knowledge base, and the probability prediction information of each text symbol is expressed as,

[0028] p i =[p i1 ,p i2 ,…,p ij ;

[0029] where j represents the number of text symbol categories, and p ij represents the probability that the text symbol belongs to a certain category.

[0030] Preferably, in step 4, combine the probability prediction information and the relationship between symbols to predict the probability sequence that each text symbol in the text symbol sequence belongs to a constant, expressed as,

[0031] [pb1,pb2,…,pb i =BSNN([p1,p2,…p l );

[0032] where l represents the length of the text symbol sequence, and pbi Indicates the probability that the symbol in the i-th position belongs to a constant.

[0033] As a second aspect of the present application, the present invention also discloses a fast abductive learning device for structured text recognition, including:

[0034] An acquisition unit for acquiring the original abductive learning architecture and knowledge base to be adjusted;

[0035] A reconstruction unit for adding machine learning model prediction probability prediction information and retaining the relationship between symbols to reconstruct the original abductive learning architecture;

[0036] A learning unit for predicting the relationship between probability prediction information and symbols to obtain a probability sequence belonging to constants;

[0037] An inference unit for converting the probability sequence belonging to constants into a discrete Boolean sequence for logical inference used in logical inference.

[0038] As a third aspect of the present application, the present invention also discloses an electronic device, including:

[0039] At least one processor, and a memory communicatively connected to the at least one processor;

[0040] Instructions executable by the at least one processor are stored on the memory, and the instructions are executed by the at least one processor so that the at least one processor can execute the steps of the above-mentioned processing method for quickly browsing common items and drawing files and merging and displaying multiple windows.

[0041] As a fourth aspect of the present application, the present invention also discloses a computer storage medium, on which a computer program is stored, characterized in that the steps of the above-mentioned processing method for quickly browsing common items and drawing files and merging and displaying multiple windows are realized when the computer program is executed by a processor.

[0042] Compared with the prior art, the beneficial effects of the present invention are:

[0043] Based on the basic framework of abductive learning, the present invention provides a fast abductive learning method and device for structured text recognition, alleviating the algorithmic deficiencies in perceptual information, text symbol relationships, and experience accumulation in the past. Using the text symbol sequence in the knowledge base learned by the machine learning model, the text symbol with the highest probability is selected as the probability prediction information, and then the neural network model is used to predict the relationship between the symbols in the text symbol sequence. Combining the probability prediction information and the relationship between the symbols to obtain the probability sequence belonging to the constant, and then generating a discrete Boolean sequence based on the probability sequence belonging to the constant to calculate the probability ranking, and selecting the discrete Boolean sequence with the highest probability to input into the obtained original abductive learning architecture for logical reasoning. Additionally, after logical reasoning, the Boolean variable with a high consistency between the knowledge base and the discrete Boolean sequence is used as the supervision information to input into the machine learning model for updating until convergence and output. The present invention endows abductive learning with the meta-reasoning ability based on text symbols similar to human experts, and can achieve remarkable results with only a small number of trial-and-error attempts. In addition, the present invention also introduces probability as an interaction interface between the conversion of continuous variables and discrete variables, converting the continuous probability sequence into a discrete Boolean sequence, which can make the fast abductive learning for structured text recognition more efficient and rigorous. It can also effectively reduce the number of runs of the logical reasoning module in abductive learning, significantly improving the efficiency of abductive learning. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] The accompanying drawings, which form a part of this application, are used to provide a further understanding of this application, making other features, objectives, and advantages of this application more apparent. The schematic embodiments and descriptions of the accompanying drawings of this application are used to explain this application and do not constitute an improper limitation of this application.

[0045] In the accompanying drawings:

[0046] Figure 1 is the step connection diagram of the abductive learning method for structured text recognition in the embodiment of the present invention;

[0047] Figure 2 is the flowchart of the abductive learning method for structured text recognition in the embodiment of the present invention;

[0048] Figure 3 is the step flowchart block diagram of the abductive learning method for structured text recognition in the embodiment of the present invention;

[0049] Figure 4 is the step flowchart block diagram of generating a discrete Boolean sequence in the embodiment of the present invention;

[0050] Figure 5 is the structural schematic diagram of the abductive learning device for structured text recognition in the embodiment of the present invention;

[0051] Figure 6Schematic diagram of the structure of the electronic device in the embodiment of the present invention. Detailed implementation manners

[0052] Embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although some embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure can be implemented in various forms and should not be construed as limited to the embodiments set forth herein. On the contrary, these embodiments are provided to more thoroughly and completely understand the present disclosure. It should be understood that the drawings and embodiments of the present disclosure are only for exemplary purposes and are not used to limit the protection scope of the present disclosure.

[0053] In addition, it should be noted that for the sake of convenience of description, only parts related to the relevant invention are shown in the drawings. Without conflict, the embodiments in the present disclosure and the features in the embodiments can be combined with each other.

[0054] The present invention discloses a fast abductive learning method and device for structured text recognition. Since the abductive efficiency of the zero-order optimization algorithm used for the conversion between digital induction and symbolic deduction is low, a fast abductive algorithm for structured text sequence recognition based on probabilistic symbol perception is proposed. The present disclosure will be described in detail below with reference to the drawings and in combination with embodiments.

[0055] Referring to Figures 1 to 3 as shown, it mainly includes the following steps:

[0056] Step 1, obtain the original abductive learning architecture and knowledge base to be adjusted;

[0057] Step 2, use a machine learning model to predict the probability of the text symbol sequence in the knowledge base to obtain the probability prediction information of the text symbol;

[0058] Step 3, use a neural network model to predict the relationship between symbols in the text symbol sequence;

[0059] Step 4, use the machine learning model again to predict the probability sequence belonging to constants in combination with the probability prediction information and the relationship between symbols;

[0060] Step 5, generate a discrete Boolean sequence based on the probability sequence belonging to constants and input it into the original abductive learning architecture for logical reasoning.

[0061] First, we need to obtain the original anti-deductive learning architecture and knowledge base for the transformation to be adjusted. Among them, the anti-deductive learning architecture and training method are known for subsequent use in restoring model performance. Generally, the anti-deductive learning architecture includes a data perception layer, a symbolic reasoning layer, and an anti-deductive reasoning engine. The data perception layer processes raw data (such as images and texts) based on a deep neural network to generate preliminary pseudo-labels or probability distributions. The symbolic reasoning layer then uses first-order logic rules to verify and correct the pseudo-labels, generating the optimal hypothesis that conforms to domain knowledge through anti-deductive reasoning. The anti-deductive reasoning engine, through the hypothesis generation-verification cycle, combines the knowledge base to screen for logically consistent hypotheses and feedbacks them to the perception layer to optimize model parameters. The training method is iterative optimization. The symbolic reasoning layer screens out logically consistent hypotheses and guides the update of the perception layer model through a loss function. The knowledge base consists of first-order logic rules, that is, the schema layer representing the knowledge base is constructed based on first-order logic rules, and entity categories, attributes, and relationship constraints are defined through an ontology library. Through the quantified expression of first-order logic, precise characterization of complex semantics can be achieved.

[0062] As described in step 2, use a machine learning model to predict the probability of the text symbol sequence in the knowledge base, and obtain the probability prediction information of each text symbol. It also includes the following steps:

[0063] Step 2.1, add the machine learning model to the original anti-deductive learning architecture;

[0064] Step 2.2, use the machine learning model to predict the probability of the text symbol sequence in the knowledge base;

[0065] Step 2.3, retain the probability prediction information of each text symbol for determining its reliability;

[0066] Step 2.4, obtain the category of the text symbol with the highest probability as the final text symbol category.

[0067] Specifically, add a machine learning model to the original anti-deductive learning architecture, and use the machine learning model to perform probability prediction on the text symbol sequence in the knowledge base. Here, the machine learning model is selected as a classification model, and all the probability prediction information results of each text symbol obtained by the classification model are retained to determine reliable prediction symbols and identify unreliable prediction symbols at the same time. Obtain the category of the text symbol with the highest probability as the final text symbol category. Among them, in all the probability prediction information during the prediction process, the probability prediction information of each text symbol is expressed as:

[0068] p i =[p i1 ,p i2 ,…,p ij ;

[0069] Among them, j represents the number of text symbol categories, p ijIndicates the probability that the text symbol belongs to a certain category.

[0070] Next, a neural network model is used to predict the relationships between symbols in the text symbol sequence. That is, the relationships between symbols in the text symbol sequence are learned through a pre-trained bidirectional sequence neural network model. In the process of understanding symbols, the meaning of a single symbol is usually limited and needs to be associated with other symbols to better understand its meaning. In addition, logical relationships are formed between multiple symbols and then extended to knowledge. Therefore, the perception at the symbol level is based on the symbol sequence. The relationships between multiple symbols can be learned using any bidirectional sequence neural network (BSNN), and the Transformer model is selected in the present invention. The Transformer model captures the global context relationship of the text through the multi-head self-attention layer. Compared with traditional RNN / CNN models, it can more accurately judge the position and type of punctuation marks. It can also avoid the semantic confusion problem caused by traditional models ignoring the word order and ensure the consistency of punctuation marks and syntactic structures.

[0071] As described in step 4, a machine learning model is further used to combine the probability prediction information and the relationships between text symbols to predict a probability sequence belonging to constants. Specifically, the machine learning model determines the category of each text symbol in the logical reasoning process, thereby obtaining a symbol probability sequence. Each value in the symbol probability sequence represents the probability of belonging to a constant in the logical reasoning process. The predicted probability of a single text symbol by the machine learning model is combined with the relationships between symbols in the symbol sequence for prediction, that is, in the logical reasoning process, it is predicted whether the category of each text symbol belongs to a variable or a constant, and the probability sequence belonging to constants is obtained and represented as

[0072] [pb1, pb2, …, pb l = BSNN([p1, p2, … p l );

[0073] where l represents the length of the text symbol sequence, and pb i represents the probability that the symbol at the i-th position belongs to a constant.

[0074] From symbol perception to symbol logical reasoning, on the basis of further considering text symbol perception, the goal of the present invention is to obtain K boolean variables for querying the knowledge base. As Figure 4 shown, specific step 5 further includes the following steps:

[0075] Step 5.1, assign values to the probability sequence belonging to constants to generate an initial discrete boolean sequence;

[0076] Step 5.2, perform a heap search based on the initial discrete boolean sequence to obtain a subsequent discrete boolean sequence;

[0077] Step 5.3, further sort the subsequent discrete Boolean sequences according to the probability;

[0078] Step 5.4, each time select the subsequent discrete Boolean sequence with the highest probability as the discrete Boolean sequence output.

[0079] Specifically, for the probability sequence belonging to constants, elements with a probability pb i greater than 0.5 are assigned 1, and vice versa are assigned 0, thus obtaining an initial discrete Boolean sequence composed of 0 and 1. Based on the initial discrete Boolean sequence, a heap search is performed to obtain the subsequent discrete Boolean sequences. When performing a heap search on the initial discrete Boolean sequence, it can be divided into three stages: heap structure construction, priority maintenance, and target screening. The execution steps of the heap search are as follows: first, quickly obtain the current extreme value through the top element of the heap, and adjust the heap structure after each element is extracted. Finally, repeat the process of extraction and adjustment until the termination condition is met. Sort the subsequent discrete Boolean sequences according to the above probability pb i further, and select the K most likely discrete Boolean sequences with the highest probability pb i as the output. Then, use the output discrete Boolean sequences to input the original abductive learning architecture for logical reasoning, which can accelerate the process of logical reasoning.

[0080] The neural network designed for processing text symbol sequences can accumulate experience during each inference process without wasting successful inference cases. Therefore, after the inference process ends, Boolean variables with a high consistency between the knowledge base and the discrete Boolean sequences during the evaluation process can be used as supervision information to update the training of the machine learning model, so that successful inference experience can be effectively accumulated. More importantly, the bidirectional sequence neural network can be pre-trained. By only considering partially replacing symbols in the originally logically correct sequence and letting the bidirectional sequence neural network predict the positions that need to be replaced, sufficient experience can be very simply accumulated before the start of abductive learning.

[0081] This embodiment is used to solve the problem of low efficiency in the logical reasoning process of abductive learning, and realizes that while reducing the reasoning cost, it can largely ensure the performance in the applicable scenarios consistent with the original model. The method and device of the present invention are mainly based on the following observations. According to existing research, the gradient-free optimization algorithm used in previous abductive learning does not take into account the advantages of combining learning and reasoning by human experts, resulting in low efficiency. There are mainly three aspects: not using the prediction information of machine learning models, not using the correlation between text symbols, and not accumulating successful abductive experience. The present invention pays attention to these three aspects of problems, and through sequence text symbol perception based on sample perception, and introducing probability as an interaction interface between the conversion of continuous variables and discrete variables, proposes an efficient and rigorous complete method for converting continuous probability sequences into discrete Boolean sequences, alleviating the above three defects.

[0082] Application of the embodiment

[0083] In this example, the effectiveness of the proposed method is verified through its application in the classical abductive learning text recognition task. The experimental design includes two specific tasks: binary addition string image recognition (DBA) and random text symbol binary addition string image recognition (RBA). At the same time, it is compared with five gradient-free optimization methods (i.e., RACOS, SRACOS, SSRACOS, POSS, PONSS) used in the logical reasoning process of classical abductive learning. In addition, different from the fact that previous abductive learning research usually allows the knowledge base to be accessed hundreds of times or more, the present invention strictly limits the number of accesses to the knowledge base to 5 times, and sets the length of a single image sample sequence to be between 5 and 10. Subsequently, according to the degree of satisfaction of each group of rules by the sample sequence and the label indicating whether it conforms to logic, a multi-layer perceptron (MLP) model is used for supervised training. In the experiment, the machine learning module uses the LeNet5 model, and the text symbol perception part uses the Transformer model, and its sequence length is set to 10. When the number of trial and error times in the logical reasoning module is limited to 5 times, the obtained accuracy rates are shown in Table 1 below. It can be seen from Table 1 that using the present invention to recognize binary addition string images and random text symbol binary addition string images has higher accuracy rates than other optimization methods.

[0084]

[0085]

[0086] Table 1

[0087] To implement the above embodiment, the present application also discloses a fast abductive learning device for structured text recognition, such as Figure 5As shown, an acquisition unit is configured to acquire an original abductive learning architecture and a knowledge base to be adjusted. A reconstruction unit is configured to add machine learning model prediction probability prediction information and retain the relationships between symbols to reconstruct the original abductive learning architecture. A learning unit is configured to predict a probability sequence belonging to constants based on the probability prediction information and the relationships between symbols. An inference unit is configured to convert the probability sequence belonging to constants into a discrete Boolean sequence for logical inference. Wherein, the inference unit further includes an acquisition module and a construction module. The reading module is configured to acquire an evaluation, determine a model architecture and a knowledge base, and the construction module is configured to generate a specified number of discrete Boolean sequences.

[0088] To implement the above embodiments, the present application also discloses an electronic device. Referring to Figure 6 As shown, the electronic device 600 may include a processing device (such as a central processing unit, a graphics processing unit, etc.) 601, which may perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 602 or a program loaded from a storage device 608 into a random access memory (RAM) 603. In the RAM 603, various programs and data required for the operation of the electronic device 600 are also stored. The processing device 601, the ROM 602, and the RAM 603 are connected to each other through a bus 604. An input / output (I / O) interface 605 is also connected to the bus 604.

[0089] Generally, the following devices may be connected to the I / O interface 605: an input device 606 including, for example, a touch screen, a touchpad, a keyboard, a mouse, a camera, a microphone, an accelerometer, a gyroscope, etc.; an output device 607 including, for example, a liquid crystal display (LCD), a speaker, a vibrator, etc.; a storage device 608 including, for example, a magnetic tape, a hard disk, etc., and a communication device 609. The communication device 609 may allow the electronic device 600 to communicate with other devices wirelessly or wiredly to exchange data. Although Figure 6 the electronic device 600 with various devices is shown, it should be understood that it is not required to implement or include all the shown devices. More or fewer devices may be alternatively implemented or included. Figure 6 Each block shown in

[0090] In particular, according to some embodiments of the present disclosure, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, some embodiments of the present disclosure include a computer program product that includes a computer program carried on a computer storage medium, and the computer program includes program code for performing the methods shown in the flowcharts. In such some embodiments, the computer program can be downloaded and installed from the network via the communication device 609, or installed from the storage device 608, or installed from the ROM 602. When the computer program is executed by the processing device 601, the above-mentioned functions defined in the methods of some embodiments of the present disclosure are performed.

[0091] It should be noted that the computer storage medium in some embodiments of the present disclosure can be a computer-readable signal medium, a computer-readable storage medium, or any combination of the two. The computer-readable storage medium can be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination of the above. More specific examples of the computer-readable storage medium can include, but are not limited to: an electrical connection having 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.

[0092] In some embodiments of the present disclosure, the computer-readable storage medium can be any tangible medium that contains or stores a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. In some embodiments of the present disclosure, the computer-readable signal medium can include a data signal propagated in a baseband or as part of a carrier wave, which carries computer-readable program code. Such a propagated data signal can take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. The computer-readable signal medium can also be any computer storage medium other than the computer-readable storage medium, and the computer-readable signal medium can send, propagate, or transmit a program for use by or in conjunction with an instruction execution system, apparatus, or device. The program code contained on the computer storage medium can be transmitted by any appropriate medium, including but not limited to: wires, optical cables, RF (radio frequency), etc., or any suitable combination of the above.

[0093] In some embodiments, the client and the server can communicate using any currently known or future-developed network protocol such as HTTP (Hyper Text Transfer Protocol), and can be interconnected with digital data communication in any form or medium (e.g., a communication network). Examples of communication networks include local area networks ("LANs"), wide area networks ("WANs"), the Internet (e.g., the Internet), and end-to-end networks (e.g., ad hoc end-to-end networks), as well as any currently known or future-developed networks.

[0094] The above computer storage medium can be included in the above electronic device; it can also exist separately without being assembled into the electronic device. The above computer storage medium carries one or more programs, and when the above one or more programs are executed by the electronic device, the electronic device can implement the processing method for quickly browsing common items and drawing files.

[0095] Computer program code for performing the operations of some embodiments of the present disclosure can be written in one or more programming languages or combinations thereof. The above programming languages include object-oriented programming languages such as Java, Smalltalk, C++, and also include conventional procedural programming languages such as the "C" language or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, executed as a stand-alone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving a remote computer, the remote computer can 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 can be connected to an external computer (e.g., by using an Internet service provider to connect through the Internet).

[0096] The flowcharts and block diagrams in the accompanying drawings illustrate the possible architectures, functions, and operations of systems, methods, and computer program products according to various embodiments of the present disclosure. In this regard, each block in the flowchart or block diagram can represent a module, a program segment, or a part of code that contains one or more executable instructions for implementing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the blocks may occur in a different order than that marked in the accompanying drawings.

[0097] For example, two consecutively represented blocks can actually be executed substantially in parallel, and they can sometimes also be executed in the reverse order, depending on the functions involved. It should also be noted that each block in the block diagram and / or flowchart, as well as the combination of blocks in the block diagram and / or flowchart, can be implemented by a dedicated hardware-based system that performs the specified functions or operations, or can be implemented by a combination of dedicated hardware and computer instructions. The units described in some embodiments of the present disclosure can be implemented in software or in hardware. The described units can also be provided in a processor, and the names of these units do not, in some cases, constitute a limitation on the unit itself.

[0098] The functions described above can be performed, at least in part, by one or more hardware logic components. For example, without limitation, exemplary types of hardware logic components that can be used include: field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems on a chip (SOCs), complex programmable logic devices (CPLDs), and so on.

[0099] The above description is only some preferred embodiments of the present disclosure and an explanation of the applied technical principles. Those skilled in the art should understand that the scope of the invention involved in the embodiments of the present disclosure is not limited to the technical solutions formed by the specific combination of the above technical features, and should also cover other technical solutions formed by any combination of the above technical features or their equivalent features without departing from the above inventive concept. For example, the technical solutions formed by mutually replacing the above features with the (but not limited to) technical features having similar functions disclosed in the embodiments of the present disclosure.

Claims

1. A fast abductive learning method for structured text recognition, characterized in that, It includes the following steps: Step 1, obtain the original abductive learning architecture and knowledge base to be adjusted; Step 2, use a machine learning model to predict the probability of the text symbol sequence in the knowledge base to obtain probability prediction information of text symbols; Step 3, use a neural network model to predict the relationship between symbols in the text symbol sequence; Step 4, use the machine learning model to combine the probability prediction information and the relationship between symbols to predict a probability sequence belonging to constants; Step 5, generate a discrete Boolean sequence based on the probability sequence belonging to constants and input it into the original abductive learning architecture for logical reasoning.

2. A fast anti-aliasing learning method for structured text recognition according to claim 1, characterized in that, Step 2 further includes the following steps: Step 2.1, add the machine learning model to the original abductive learning architecture; Step 2.2, use the machine learning model to predict the probability of the text symbol sequence in the knowledge base; Step 2.3, retain the probability prediction information of each text symbol for determining its reliability; Step 2.4, obtain the category of the text symbol with the highest probability as the final text symbol category.

3. A fast anti-aliasing learning method for structured text recognition according to claim 2, characterized in that, Step 5 further includes the following steps: Step 5.1, generate an initial discrete Boolean sequence according to the assignment of the probability sequence belonging to constants; Step 5.2, perform a heap search based on the initial discrete Boolean sequence to obtain a subsequent discrete Boolean sequence; Step 5.3, further sort the subsequent discrete Boolean sequences according to probability; Step 5.4, each time select the subsequent discrete Boolean sequence with the highest probability as the discrete Boolean sequence output.

4. A fast anti-aliasing learning method for structured text recognition according to claim 3, characterized in that: In step 5.1, generate an initial discrete Boolean sequence according to the assignment of the probability sequence belonging to constants, and its assignment rule is to assign a probability greater than 0.5 in the probability sequence belonging to constants as 1, otherwise assign 0.

5. A fast anti-aliasing learning method for structured text recognition according to claim 4, characterized in that: In step After step 5 performs logical reasoning, use the Boolean variable with higher consistency between the knowledge base and the discrete Boolean sequence as supervision information, and input the supervision information into the machine learning model for updating until convergence.

6. A fast anti-aliasing learning method for structured text recognition according to claim 5, characterized in that: In step 2, when predicting the probability of the text symbol sequence in the knowledge base, the probability prediction information of each text symbol is expressed as p i = [p i1 , p i2 , …, p ij ; where j represents the number of text symbol categories, and p ij represents the probability that the text symbol belongs to a certain category.

7. A fast anti-aliasing learning method for structured text recognition according to claim 5, characterized in that: In step 4, by combining the probability prediction information and the relationship between symbols, predict a probability sequence in which each text symbol in the text symbol sequence belongs to a constant, expressed as [pb1, pb2, …, pb i = BSNN([p1, p2, … p l ); where l represents the length of the text symbol sequence, and pb i represents the probability that the symbol at the i-th position belongs to a constant.

8. A fast abductive learning device for structured text recognition, characterized in that, It includes: An acquisition unit for acquiring the original abductive learning architecture and knowledge base to be adjusted; A reconstruction unit for adding a machine learning model to predict probability prediction information and retain the relationship between symbols to reconstruct the original abductive learning architecture; A learning unit for predicting probability prediction information and the relationship between symbols to obtain a probability sequence belonging to constants; An inference unit for converting the probability sequence belonging to constants into a discrete Boolean sequence for logical reasoning used in logical reasoning.

9. An electronic device, characterized in that, It includes: At least one processor, and a memory communicatively connected to the at least one processor; Instructions executable by the at least one processor are stored on the memory, and the instructions are executed by the at least one processor so that the at least one processor can execute the steps of the method according to any one of claims 1 to 7.

10. A computer storage medium, on which a computer program is stored, characterized in that: When the computer program is executed by a processor, the steps described in any one of claims 1 to 7 are implemented.

Images