WO2022127610A1

WO2022127610A1 - Text recognition result processing system, method and device

Info

Publication number: WO2022127610A1
Application number: PCT/CN2021/135047
Authority: WO
Inventors: 杨建国; 詹镇江
Original assignee: 第四范式（北京）技术有限公司
Priority date: 2020-12-16
Filing date: 2021-12-02
Publication date: 2022-06-23
Also published as: CN114637816A

Abstract

The present disclosure provides a text recognition result processing system, method and device. Said method comprises: acquiring a text recognition result of a text recognition model, and detecting whether a text matching the text recognition result exists in a word library; if there is no matched text, performing word segmentation on the text recognition result, to obtain a word set; according to inverted index information of each word of the word set in an inverted index of the word library, acquiring a text set matching the text recognition result; and selecting one text from the text set as a final text recognition result. The present disclosure solves the problem of low accuracy of a text recognition result in the related art.

Description

Text recognition result processing system, method and device

This disclosure claims the priority of a Chinese patent application with an application number of 202011487618.7 and an application date of December 16, 2020, titled "Text Recognition Result Processing Method, Device and Computer-readable Storage Medium", wherein the content disclosed in the above application Incorporated in this disclosure by reference.

technical field

The present disclosure relates to the field of computers, and the following description relates to a text recognition result processing system, method and apparatus.

Background technique

In production and life, people have to deal with a large number of words, reports and texts. In order to reduce people's labor and improve processing efficiency, general text recognition methods began to be discussed in the 1950s. Text recognition is divided into two specific steps: text detection and text recognition, both of which are indispensable. For text recognition, the industry has reached a consensus. The accuracy of the numbers in the text is generally high (above 95%), while for the text in the text, one type is non-open text (the range of values can be enumerated, such as capitalized dates, local city names, capitalized amount, etc.), the accuracy rate can generally be increased to more than 90%, the other type is open text (the range of values is not enumerable, such as the recognition of company names, etc.), because of the continuous increase of data & the diversity of characters, resulting in the model's The industry standard for the accuracy rate is usually 75%, and this effect cannot be applied to the production business system. The frequency of sample changes and the difficulty of self-learning make the defects of this problem increasingly obvious.

At present, the commonly used text recognition method is optical character recognition (Optical Character Recognition, abbreviated as ocr), but whether it is the traditional ocr method or the end-to-end deep learning ocr recognition network, such as crnn, crnn+ctc, seq2seq-attention, etc., It is a prerequisite for training and learning, which is a prerequisite. If this prerequisite is met, the field of open text recognition usually only reaches 75%, and many small banks or small companies are facing this problem. When there is a practical problem, there are not enough samples to support it. If only the recognition model trained with a small amount of data is used to support the business system, it will not be enough. At this time, the engineering model compensation scheme to improve the overall accuracy rate is particularly important.

There is no solution for the problem of low accuracy of text recognition results in the related art.

SUMMARY OF THE INVENTION

Exemplary embodiments of the present disclosure are to provide a text recognition result processing system, method and apparatus, which can solve the problem of low accuracy of text recognition results in the related art.

According to a first aspect of the present disclosure, there is provided a system comprising at least one computing device and at least one storage device storing instructions, wherein the instructions, when executed by the at least one computing device, cause the at least one computing device to The following method of executing the text recognition result processing method: obtain the text recognition result of the text recognition model, and detect whether there is text matching the text recognition result in the thesaurus; when there is no matching text, cut the text recognition result. word to obtain a word set; according to the inverted index information of each word in the word set in the inverted index of the thesaurus, a text set matching the text recognition result is obtained; a text is selected from the text set as the final text recognition result.

According to a second aspect of the present disclosure, there is provided a method for processing a text recognition result, the method comprising: acquiring a text recognition result of a text recognition model, and detecting whether there is text matching the text recognition result in the thesaurus; In the case of text, segment the text recognition result to obtain a word set; according to the inverted index information of each word in the word set in the inverted index of the thesaurus, obtain a text set that matches the text recognition result; Select a text in the collection as the final text recognition result.

According to a third aspect of the present disclosure, a text recognition result processing device includes: a storage unit for storing a thesaurus and an inverted index of words in the thesaurus; a compensation processing unit for acquiring a text recognition model Text recognition results, and detect whether there is text matching the text recognition results in the thesaurus; when there is no matching text, segment the text recognition results to obtain a word set; The inverted index information in the inverted index of , obtains a text set that matches the text recognition result; selects a text from the text set as the final text recognition result.

According to a fourth aspect of the present disclosure, there is provided a computer-readable storage medium storing instructions, wherein the instructions, when executed by at least one computing device, cause the at least one computing device to perform the text recognition result processing as described above method.

According to a fifth aspect of the present disclosure, there is provided an electronic device, comprising: a processor; a memory for storing instructions executable by the processor; wherein the processor is configured to execute the instructions to implement the present disclosure The text recognition result processing method.

According to the text recognition result processing system, method, and device of the present exemplary embodiment, when the text recognition result obtained by the text recognition model is not in the memory, the recognition result is segmented, and a suitable word is matched according to the inverted index information of the segmented words. A text set, from which the final text recognition result fed back to the customer is determined, so that the text recognition result fed back to the customer is more accurate, the accuracy of the recognition result is improved, and the low accuracy of the text recognition result in related technologies is solved. The problem.

Description of drawings

These and/or other aspects and advantages of the present disclosure will become apparent, and be more readily understood, from the following description of embodiments, taken in conjunction with the accompanying drawings, wherein:

1 shows a flowchart of a method for processing a text recognition result according to an exemplary embodiment of the present disclosure;

2 shows a schematic flowchart of a preferred text recognition result processing method according to an exemplary embodiment of the present disclosure;

FIG. 3 shows an overall architecture diagram of a method for processing a text recognition result according to an exemplary embodiment of the present disclosure;

FIG. 4 shows a structural block diagram of a text recognition result processing apparatus according to an exemplary embodiment of the present disclosure.

Detailed ways

The following description with reference to the accompanying drawings is provided to assist in a comprehensive understanding of embodiments of the present disclosure as defined by the claims and their equivalents. Various specific details are included to aid in that understanding, but are to be regarded as exemplary only. Accordingly, those of ordinary skill in the art will recognize that various changes and modifications of the embodiments described herein can be made without departing from the scope and spirit of the present disclosure. Also, descriptions of well-known functions and constructions are omitted for clarity and conciseness.

It should be noted here that "at least one of several items" in the present disclosure all means including "any one of the several items", "a combination of any of the several items", The three categories of "the whole of the several items" are juxtaposed. For example, "including at least one of A and B" includes the following three parallel situations: (1) including A; (2) including B; (3) including A and B. Another example is "execute at least one of step 1 and step 2", which means the following three parallel situations: (1) execute step 1; (2) execute step 2; (3) execute step 1 and step 2. That is to say, "A and/or B" can also be expressed as "at least one of A and B", and "execute step 1 and/or step 2" can also be expressed as "execute step 1 and step 2" at least one of".

Reference will now be made in detail to the embodiments of the present disclosure, examples of which are illustrated in the accompanying drawings. The embodiments are described below in order to explain the present disclosure by referring to the figures.

FIG. 1 shows a flowchart of a text recognition result processing method according to an exemplary embodiment of the present disclosure.

1, in step S101, the text recognition result of the text recognition model is obtained, and it is detected whether there is a text matching the text recognition result in the thesaurus;

In an embodiment of the present disclosure, before acquiring the text recognition result of the text recognition model, the method further includes: detecting that the text recognition service is started, maintaining the thesaurus and the inverted index of the words in the thesaurus in the buffer memory; detecting Whether there is text matching the text recognition result in the thesaurus includes: detecting whether there is text matching the text recognition result in the buffer memory. Through this embodiment, the inverted index of the word set is maintained in the cache, and the query from the cache is faster.

In an embodiment of the present disclosure, before maintaining the words in the thesaurus and the inverted index corresponding to the words in the buffer memory, the method further includes: acquiring the thesaurus; segmenting all the texts in the thesaurus; acquiring the segmented words Inverted index information for each subsequent word. Through this embodiment, the inverted index information of the words is acquired in advance, and the text recognition result can be directly queried, which saves processing time.

It should be noted that an inverted index (Inverted index), also often referred to as an inverted index, an inserted file or a reversed file, is an indexing method that is used to store a word in a document under full-text search. Or a map of storage locations within a set of documents. It is the most commonly used data structure in document retrieval systems. With an inverted index, you can quickly get a list of documents that contain a word based on that word.

It should be noted that, the above-mentioned thesaurus can be obtained in a targeted manner according to the business type of the text to be recognized by the text recognition service. For example, if the text is a banking business, a thesaurus related to the banking business is obtained.

In an embodiment of the present disclosure, the above-mentioned word segmentation for all texts in the thesaurus can be performed by using a search engine mode word segmentation method to segment the text recognition results, wherein the search engine mode word segmentation method will Long words are further divided into secondary words.

In step S102, when there is no matching text, the text recognition result is segmented to obtain a word set;

In an embodiment of the present disclosure, the above-mentioned word segmentation of the text recognition result can be performed by using the precise mode word segmentation method to segment the text recognition result, wherein the precise mode word segmentation method does not further segment the long words Secondary participle.

In step S103, according to the inverted index information of each word in the word set in the inverted index of the thesaurus, obtain a text set that matches the text recognition result;

In an embodiment of the present disclosure, according to the inverted index information of each word in the word set in the inverted index of the thesaurus, acquiring a text set that matches the text recognition result includes: querying the word set for each word in the word set The inverted index information in the inverted index of the library; according to the inverted index information, obtain the text identification set of the text matching each word; determine the number of occurrences of each text identification in the text identification set; The text corresponding to the text identifier is merged into a text set matching the text recognition result. With this embodiment, some irrelevant texts are removed first, which greatly reduces the amount of texts that need to be calculated for the edit distance.

In step S104, a text is selected from the text set as the final text recognition result.

In an embodiment of the present disclosure, selecting a text from the text set as the final text recognition result includes: obtaining an edit distance between each text in the text set and the text recognition result; Minimum edit distance; determine the text corresponding to the minimum edit distance as the final text recognition result.

In an embodiment of the present disclosure, after selecting a text from the text set as the final text recognition result, the method further includes: sending the final text recognition result to the client.

In an embodiment of the present disclosure, after sending the final text recognition result to the client, the method further includes: receiving an error correction request sent by the client based on the feedback final text recognition result, wherein the error correction request carries The correct text corresponding to the final text recognition result with feedback; the correct text is stored in the buffer memory. Through this embodiment, when the client determines that the returned text recognition result is wrong, the correct result is stored in the buffer memory, which ensures the accuracy of the wrong text recognition result in the future without relying on frequent updates of the model , and solve a class of model self-learning closed-loop difficult problems through engineering closed-loop.

In an embodiment of the present disclosure, the above method further includes: in the case where it is detected that the acquired text recognition result matches the corresponding text, feeding back the text corresponding to the text recognition result to the client. With this embodiment, when the recognized text is in the memory, it can be directly fed back to the client.

To sum up, in view of the problem that the recognition accuracy of the text recognition model is not high, the text recognition result processing method of this exemplary embodiment, after acquiring the text recognition result of the text recognition model, performs a cache hit judgment on the text recognition result. , that is, to determine whether the text recognition result is in the existing thesaurus, and if it is in the existing thesaurus, return it immediately. If there is no word in the thesaurus, perform precise pattern segmentation on the text recognition result, and calculate the text set A in the thesaurus that best matches the text recognition result based on the inverted index information of each word after word segmentation. After this calculation, The set of phrases to be searched is greatly reduced, and then by calculating the edit distance between the text recognition result and each text in the text set A, the text corresponding to the smallest distance is theoretically the one result that the model should return most, and the text corresponding to the smallest distance is calculated. The text is fed back to the client, which greatly improves the accuracy of text recognition results. Furthermore, by introducing the buffer memory cache & Chinese word segmentation, it is possible to calculate massive edit distances in real time.

Aiming at the problem of lack of samples, slow accumulation and difficulty in self-learning, the text recognition result processing method of this exemplary embodiment adds the design of correct result (label) feedback. When the client determines that the returned text recognition result is wrong, Then, the correct result is sent to the buffer memory, specifically, it can be sent to the designed backflow interface, and the solution of the engineering end can ensure the accuracy of the same error text in the future, without relying on frequent updates of the model.

The following takes the text recognition result as "I came to Beijing Qing University" as an example to describe in detail how the text recognition result processing method of the present disclosure solves the problem of low model recognition accuracy.

As shown in Figure 2, first, the online recognition model recognizes the text recognition result "I came to Beijing Qing University", and makes a cache hit judgment for "I came to Beijing Qing University". If it is determined to be in the existing hash structure Hash, Return immediately (ie hit). If there is no "I came to Beijing Qing University" in the Hash, then "I came to Beijing Qing University" will be segmented, and the set of words "I", "Come", "Beijing", "Qing" and "University" will be obtained. The inverted index information of each word after the word, the text identifier corresponding to the text matched by each word is obtained, and the number of occurrences of each text identifier is calculated. The text set A in the thesaurus that most matches the text recognition result, and then by calculating the edit distance between the text recognition result and each text in the text set A, the text corresponding to the smallest distance is fed back to the client.

It should be noted that, based on the virtual hash slot partition storage & operator design of redis-cluster cluster, it is possible to configure some parameters of the operator to control whether to intervene in the recognition model results, and also to achieve a large number of phrases. real-time compensation.

The exemplary embodiment of the present disclosure can be divided into three parts from bottom to top. The first part is the cold start stage, preparing a historical thesaurus, performing search mode segmentation on all phrases in the thesaurus, and calculating the inverse of all the segmented words. Sort index information. The second part is the online service stage. When the service is started, the thesaurus and the inverted index information of the first part are maintained in the redis-cluster (or other cache structures), and the online model recognition results are accurately modeled. Calculate the set of phrases that best match the word segmentation, and then calculate the edit distance (Levenshtein Distance) between the recognition result and these phrases. The third part is the label result feedback. When the returned result is judged wrong in the client system, the correct result is fed back to the system.

The following is a detailed description in conjunction with Figure 3. As shown in Figure 3, the entire solution can be divided into three parts, the cold start stage, the online service & label feedback stage. The overall structure is as follows:

1. In the cold start stage, prepare the thesaurus file, and generate the corresponding inverted index file through the word segmentation script.

2. When the service is started, load thesaurus file & inverted index file into the redis-cluster cache. In order to improve the query effect, thesaurus files can be stored redundantly, for example, thesaurus information is stored in the id dimension and the name dimension respectively.

3. When requesting an online service, first obtain the model recognition result, and then check whether it exists through the name dimension. If it exists, it will return to the client immediately. Otherwise, the recognition result will be segmented, and the maximum matching phrase id set will be calculated through the inverted index information. Query the phrase set in the id dimension, calculate the phrase with the shortest distance through the edit distance, and return it to the client.

4. If the user judges that the returned result is still wrong, the correct label corresponding to the request needs to be fed back to the designed reflow interface.

FIG. 4 shows a structural block diagram of a text recognition result processing apparatus according to an exemplary embodiment of the present disclosure. As shown in Figure 4, the device includes:

The storage unit 40 is used to store the thesaurus and the inverted index of the words in the thesaurus;

The compensation processing unit 42 is used to obtain the text recognition result of the text recognition model, and detects whether there is a text matching the text recognition result in the thesaurus; when there is no matching text, the text recognition result is segmented to obtain words Set; according to the inverted index information of each word in the word set in the inverted index of the thesaurus, obtain a text set that matches the text recognition result; select a text from the text set as the final text recognition result.

In an embodiment of the present disclosure, the compensation processing unit 42 is further configured to query the inverted index information of each word in the word set in the inverted index of the thesaurus; according to the inverted index information, obtain a match for each word The text identification set of the text; determine the number of occurrences of each text identification in the text identification set; merge the texts corresponding to the text identifications whose frequency exceeds a predetermined number of times into a text set matching the text recognition result.

In an embodiment of the present disclosure, the compensation processing unit 42 is further configured to obtain the edit distance between each text in the text set and the text recognition result; sort the edit distances to obtain the minimum edit distance among the edit distances; The text corresponding to the distance is determined as the final text recognition result.

Optionally, the compensation processing unit 42 is further configured to detect that the text recognition service is started before acquiring the text recognition result of the text recognition model, and maintain the thesaurus and the inverted index of the words in the thesaurus in the buffer memory; detecting Whether there is text matching the text recognition result in the buffer memory.

In an embodiment of the present disclosure, the compensation processing unit is further configured to send the final text recognition result to the client after selecting a text from the text set as the final text recognition result.

In an embodiment of the present disclosure, the compensation processing unit 42 is further configured to, after sending the final text recognition result to the client, receive an error correction request sent by the client based on the feedback final text recognition result, wherein the correction The error request carries the correct text corresponding to the feedback final text recognition result; the correct text is stored in the buffer memory.

In an embodiment of the present disclosure, the compensation processing unit 42 is further configured to obtain the thesaurus before maintaining the words in the thesaurus and the inverted index corresponding to the words in the buffer memory; cut all the texts in the thesaurus word; obtains the inverted index information of each word after word segmentation.

In an embodiment of the present disclosure, the compensation processing unit 42 is further configured to feed back the text corresponding to the text recognition result to the client when it is detected that the acquired text recognition result matches the corresponding text.

The present disclosure constructs an engineering solution for improving the low accuracy rate of an open text recognition model, which is used to solve the problem in the ocr field, such as the situation where the accuracy rate of some similar fields in bank bill recognition cannot meet the standard, and the industry is in the open text The accuracy of the recognition model is usually around 75%. This kind of problem usually cannot be self-learned by automatically acquiring a large number of samples and updating the model in real time to improve the accuracy. With the engineering solutions of the present disclosure, the accuracy rate can typically be improved to over 90%.

The method and apparatus for processing a text recognition result according to an exemplary embodiment of the present disclosure have been described above with reference to FIGS. 1 to 4 .

Each unit in the text recognition result processing apparatus shown in FIG. 4 may be configured as software, hardware, firmware or any combination of the above items to perform specific functions. For example, each unit may correspond to a dedicated integrated circuit, may also correspond to a pure software code, or may correspond to a module combining software and hardware. In addition, one or more functions implemented by each unit can also be uniformly performed by components in a physical entity device (eg, a processor, a client or a server, etc.).

Furthermore, the text recognition result processing method described with reference to FIG. 1 may be implemented by a program (or instruction) recorded on a computer-readable storage medium. For example, in accordance with exemplary embodiments of the present disclosure, a computer-readable storage medium storing instructions may be provided that, when executed by at least one computing device, cause the at least one computing device to perform assisted labor according to the present disclosure Method for text annotation.

The computer program in the above-mentioned computer-readable storage medium can run in an environment deployed in computer equipment such as a client, a host, an agent device, a server, etc. It should be noted that the computer program can also be used to perform additional steps in addition to the above-mentioned steps or More specific processing is performed when the above steps are performed, and the contents of these additional steps and further processing have been mentioned in the description of the related method with reference to FIG. 1 , and thus will not be repeated here to avoid repetition.

It should be noted that each unit in the text recognition result processing apparatus according to the exemplary embodiment of the present disclosure can completely rely on the running of the computer program to realize the corresponding function, that is, each unit corresponds to each step in the functional architecture of the computer program, so that The entire system is invoked through specialized software packages (eg, lib libraries) to implement corresponding functions.

On the other hand, each unit shown in FIG. 4 can also be implemented by hardware, software, firmware, middleware, microcode or any combination thereof. When implemented in software, firmware, middleware, or microcode, program codes or code segments for performing corresponding operations may be stored in a computer-readable medium such as a storage medium, so that a processor can read and execute the corresponding program by reading code or code segment to perform the corresponding action.

For example, exemplary embodiments of the present disclosure may also be implemented as a computing device including a storage component and a processor, the storage component stores a computer-executable instruction set, and when the computer-executable instruction set is executed by the processor, executes the A text recognition result processing method according to an exemplary embodiment of the present disclosure.

Specifically, the computing device may be deployed in a server or a client, or may be deployed on a node device in a distributed network environment. Furthermore, the computing device may be a PC computer, a tablet device, a personal digital assistant, a smartphone, a web application, or other device capable of executing the set of instructions described above.

Here, the computing device does not have to be a single computing device, but can also be any set of devices or circuits capable of individually or jointly executing the above-mentioned instructions (or instruction sets). The computing device may also be part of an integrated control system or system manager, or may be configured as a portable electronic device that interfaces locally or remotely (eg, via wireless transmission).

In a computing device, a processor may include a central processing unit (CPU), a graphics processing unit (GPU), a programmable logic device, a special purpose processor system, a microcontroller, or a microprocessor. By way of example and not limitation, processors may also include analog processors, digital processors, microprocessors, multi-core processors, processor arrays, network processors, and the like.

Some operations described in the text recognition result processing method according to an exemplary embodiment of the present disclosure may be implemented by software, some operations may be implemented by hardware, and in addition, these operations may also be implemented by a combination of software and hardware operate.

The processor may execute instructions or code stored in one of the storage components, which may also store data. Instructions and data may also be sent and received over a network via a network interface device, which may employ any known transport protocol.

The memory component may be integrated with the processor, eg, RAM or flash memory arranged within an integrated circuit microprocessor or the like. Additionally, the storage components may include separate devices, such as external disk drives, storage arrays, or any other storage device that may be used by a database system. The storage component and the processor may be operatively coupled, or may communicate with each other, eg, through I/O ports, network connections, etc., to enable the processor to read files stored in the storage component.

In addition, the computing device may also include a video display (such as a liquid crystal display) and a user interaction interface (such as a keyboard, mouse, touch input device, etc.). All components of the computing device may be connected to each other via a bus and/or network.

The text recognition result processing method according to an exemplary embodiment of the present disclosure may be described as various interconnected or coupled functional blocks or functional diagrams. However, these functional blocks or functional diagrams may be equally integrated into a single logical device or operate along non-precise boundaries.

Therefore, the text recognition result processing method described with reference to FIG. 1 can be implemented by a system including at least one computing device and at least one storage device storing instructions.

According to an exemplary embodiment of the present disclosure, at least one computing device is a computing device for executing a method for processing a text recognition result according to an exemplary embodiment of the present disclosure, and a computer-executable instruction set is stored in the storage device. When the collection is executed by at least one computing device, the text recognition result processing method described with reference to FIG. 1 is executed.

According to an exemplary embodiment of the present disclosure, there is provided an electronic device, comprising: a processor; a memory for storing instructions executable by the processor; wherein the processor is configured to execute the instructions to implement a reference Figure 1 describes the text recognition result processing method.

Various exemplary embodiments of the present disclosure have been described above, and it should be understood that the above description is merely exemplary and not exhaustive, and the present disclosure is not limited to the disclosed exemplary embodiments. Numerous modifications and variations will be apparent to those of ordinary skill in the art without departing from the scope and spirit of this disclosure. Therefore, the scope of protection of the present disclosure should be determined by the scope of the claims.

Industrial Applicability

According to the text recognition result processing method of the present disclosure, when the text recognition result obtained by the text recognition model is not in the memory, the recognition result is segmented, and a suitable text set is matched according to the inverted index information of the segmented words, and the text set is obtained from the text set. The final text recognition result fed back to the customer is determined in the middle of the paper, which makes the text recognition result fed back to the customer more accurate, improves the accuracy of the recognition result, and solves the problem of low accuracy of the text recognition result in related technologies.

Claims

A system comprising at least one computing device and at least one storage device storing instructions, wherein the instructions, when executed by the at least one computing device, cause the at least one computing device to perform the following steps of a text recognition result processing method :

Obtain the text recognition result of the text recognition model, and detect whether there is text matching the text recognition result in the thesaurus;

When there is no matching text, segment the text recognition result to obtain a word set;

According to the inverted index information of each word in the word set in the inverted index of the thesaurus, obtain a text set that matches the text recognition result;

One text is selected from the text set as the final text recognition result.
The system according to claim 1, wherein, according to the inverted index information of each word in the word set in the inverted index of the thesaurus, acquiring a text set matching the text recognition result include:

query the inverted index information of each word in the word set in the inverted index of the thesaurus;

According to the inverted index information, obtain a text identification set of the text matching each word;

Determining the number of occurrences of each text identifier in the text identifier set;

The texts corresponding to the text identifiers whose number of times exceeds a predetermined number of times are combined into a text set matching the text recognition result.
The system according to claim 1, wherein the selecting a text from the text set as the final text recognition result comprises:

Obtain the edit distance between each text in the text set and the text recognition result;

Sorting the edit distances to obtain the minimum edit distance among the edit distances;

The text corresponding to the minimum edit distance is determined as the final text recognition result.
The system of claim 1, wherein,

When executed by the at least one computing device, the instruction further causes the at least one computing device to perform the following steps: before acquiring the text recognition result of the text recognition model, detecting that the text recognition service is started, and converting the thesaurus to the text recognition service. maintaining the inverted index of the words in the thesaurus in the buffer memory;

The detecting whether the text matching the text recognition result exists in the thesaurus includes: detecting whether the text matching the text recognition result exists in the buffer memory.
5. The system of claim 4, wherein the instructions, when executed by the at least one computing device, further cause the at least one computing device to perform the step of selecting a text from the set of texts as a final After the text recognition result is obtained, the final text recognition result is sent to the client.
6. The system of claim 5, wherein the instructions, when executed by the at least one computing device, further cause the at least one computing device to perform the step of: after sending the final text recognition result to the client ,

receiving an error correction request sent by the client based on the feedback final text recognition result, wherein the error correction request carries the correct text corresponding to the feedback final text recognition result;

The correct text is stored in the buffer memory.
5. The system of claim 4, wherein the instructions, when executed by the at least one computing device, further cause the at least one computing device to perform the step of: comparing the terms in the thesaurus with the Obtain the thesaurus before the inverted index corresponding to the word is maintained in the buffer memory;

segmenting all texts in the thesaurus;

Obtain the inverted index information of each word after word segmentation.
7. The system of any one of claims 1 to 7, wherein the instructions, when executed by the at least one computing device, further cause the at least one computing device to perform the following steps:

When it is detected that the acquired text recognition result matches the corresponding text, the text corresponding to the text recognition result is fed back to the client.
A text recognition result processing method, wherein the method comprises:

Obtain the text recognition result of the text recognition model, and detect whether there is text matching the text recognition result in the thesaurus;

When there is no matching text, word segmentation is performed on the text recognition result to obtain a word set;

According to the inverted index information of each word in the word set in the inverted index of the thesaurus, acquiring a text set that matches the text recognition result;

One text is selected from the text set as the final text recognition result.
The method according to claim 9, wherein, according to the inverted index information of each word in the word set in the inverted index of the thesaurus, acquiring a text set matching the text recognition result include:

query the inverted index information of each word in the word set in the inverted index of the thesaurus;

According to the inverted index information, obtain a text identification set of the text matching each word;

Determining the number of occurrences of each text identifier in the text identifier set;

The texts corresponding to the text identifiers whose number of times exceeds a predetermined number of times are combined into a text set matching the text recognition result.
The method according to claim 9, wherein the selecting a text from the text set as the final text recognition result comprises:

Obtain the edit distance between each text in the text set and the text recognition result;

Sorting the edit distances to obtain the minimum edit distance among the edit distances;

The text corresponding to the minimum edit distance is determined as the final text recognition result.
The method of claim 9, wherein,

Before acquiring the text recognition result of the text recognition model, the method further includes: detecting that the text recognition service is started, and maintaining the thesaurus and the inverted index of the words in the thesaurus in the buffer memory;

The detecting whether the text matching the text recognition result exists in the thesaurus includes: detecting whether the text matching the text recognition result exists in the buffer memory.
The method according to claim 12, wherein after selecting a text from the text set as the final text recognition result, the method further comprises: sending the final text recognition result to the client.
The method according to claim 13, wherein after sending the final text recognition result to the client, the method further comprises:

receiving an error correction request sent by the client based on the feedback final text recognition result, wherein the error correction request carries the correct text corresponding to the feedback final text recognition result;

The correct text is stored in the buffer memory.
The method according to claim 12, wherein before maintaining the words in the thesaurus and the inverted index corresponding to the words in the buffer memory, the method further comprises:

get thesaurus;

segmenting all texts in the thesaurus;

Obtain the inverted index information of each word after word segmentation.
The method of any one of claims 9 to 15, wherein the method further comprises:

When it is detected that the acquired text recognition result matches the corresponding text, the text corresponding to the text recognition result is fed back to the client.
A text recognition result processing device, wherein the device comprises:

a storage unit for storing a thesaurus and an inverted index of the words in the thesaurus;

a compensation processing unit, configured to obtain the text recognition result of the text recognition model, and detect whether there is text matching the text recognition result in the thesaurus; when there is no matching text, the text recognition result Perform word segmentation to obtain a word set; according to the inverted index information of each word in the word set in the inverted index of the thesaurus, obtain a text set that matches the text recognition result; from the text set Choose a text as the final text recognition result.
A computer-readable storage medium storing instructions, wherein the instructions, when executed by at least one computing device, cause the at least one computing device to perform the text recognition of any one of claims 9-16 Result processing method.
An electronic device comprising:

processor;

memory for storing instructions executable by the processor;

Wherein, the processor is configured to execute the instructions to implement the text recognition result processing method according to any one of claims 9 to 16.