WO2021000412A1 - 文本匹配度检测方法、装置、计算机设备和可读存储介质 - Google Patents

文本匹配度检测方法、装置、计算机设备和可读存储介质 Download PDF

Info

Publication number
WO2021000412A1
WO2021000412A1 PCT/CN2019/103474 CN2019103474W WO2021000412A1 WO 2021000412 A1 WO2021000412 A1 WO 2021000412A1 CN 2019103474 W CN2019103474 W CN 2019103474W WO 2021000412 A1 WO2021000412 A1 WO 2021000412A1
Authority
WO
WIPO (PCT)
Prior art keywords
target
self
text
encoding structure
encoding
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Ceased
Application number
PCT/CN2019/103474
Other languages
English (en)
French (fr)
Inventor
金戈
徐亮
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Ping An Technology Shenzhen Co Ltd
Original Assignee
Ping An Technology Shenzhen Co Ltd
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Ping An Technology Shenzhen Co Ltd filed Critical Ping An Technology Shenzhen Co Ltd
Publication of WO2021000412A1 publication Critical patent/WO2021000412A1/zh
Anticipated expiration legal-status Critical
Ceased legal-status Critical Current

Links

Images

Classifications

    • GPHYSICS
    • G06COMPUTING OR CALCULATING; COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/30Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
    • G06F16/33Querying
    • G06F16/3331Query processing
    • G06F16/334Query execution
    • G06F16/3343Query execution using phonetics
    • GPHYSICS
    • G06COMPUTING OR CALCULATING; COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/30Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
    • G06F16/33Querying
    • G06F16/3331Query processing
    • G06F16/334Query execution
    • G06F16/3344Query execution using natural language analysis
    • GPHYSICS
    • G06COMPUTING OR CALCULATING; COUNTING
    • G06NCOMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00Computing arrangements based on biological models
    • G06N3/02Neural networks
    • G06N3/04Architecture, e.g. interconnection topology
    • G06N3/045Combinations of networks
    • GPHYSICS
    • G06COMPUTING OR CALCULATING; COUNTING
    • G06NCOMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00Computing arrangements based on biological models
    • G06N3/02Neural networks
    • G06N3/08Learning methods

Definitions

  • This application relates to a text matching degree detection method, device, computer equipment and readable storage medium.
  • Text matching degree refers to the degree of semantic relevance between different texts.
  • the determination of text matching degree is one of the core tasks of text mining and text retrieval; for example, when using speech keywords to retrieve relevant documents from a database, text matching is required Degree detection and corresponding documents of voice keywords. Therefore, how to better perform text matching degree detection has always been an issue of great concern to those skilled in the art.
  • the main method for detecting text matching degree in the prior art is: mapping the text into a vector in the word space, and calculating the Euclidean distance or the cosine distance between the vectors.
  • the inventor realizes that the existing detection method cannot effectively extract the semantic features of the text, resulting in a relatively low detection accuracy rate of the text matching degree, resulting in a low accuracy rate of the matching degree detection result.
  • a text matching degree detection method is provided.
  • a text matching degree detection method includes:
  • the target text is input into a preset target self-encoding structure, and the implicit feature vector corresponding to the target text is extracted from the intermediate layer of the target self-encoding structure; wherein, the target self-encoding structure is used to Encoding and decoding the target text to obtain the hidden feature vector; the hidden feature vector is used to represent the semantic feature of the target text;
  • the similarity is taken as the matching degree between the target text and the target standard text.
  • a text matching degree detection device includes:
  • the text acquisition module acquires the input voice text, and converts the voice text into target text
  • the hidden feature acquisition module is used to input the target text into a preset target self-encoding structure, and extract the hidden feature vector corresponding to the target text from the middle layer of the target self-encoding structure; wherein, the target The self-encoding structure is used to encode and decode the target text to obtain the hidden feature vector; the hidden feature vector is used to represent the semantic feature of the target text;
  • the standard vector obtaining module is configured to obtain a preset target standard text corresponding to the target text, and convert the target standard text into a target standard vector;
  • the matching degree detection module is used to obtain the similarity between the hidden feature vector and the target standard vector, according to the correspondence between the hidden feature vector and the target text, and the target standard vector For the correspondence relationship with the target standard text, the similarity is taken as the matching degree between the target text and the target standard text.
  • a computer device including a memory and one or more processors, the memory stores computer readable instructions, when the computer readable instructions are executed by the processor, the one or more processors execute The following steps:
  • the target text is input into a preset target self-encoding structure, and the implicit feature vector corresponding to the target text is extracted from the intermediate layer of the target self-encoding structure; wherein, the target self-encoding structure is used to Encoding and decoding the target text to obtain the hidden feature vector; the hidden feature vector is used to represent the semantic feature of the target text;
  • the similarity is taken as the matching degree between the target text and the target standard text.
  • One or more non-volatile computer-readable storage media storing computer-readable instructions.
  • the one or more processors execute the following steps:
  • the target text is input into a preset target self-encoding structure, and the implicit feature vector corresponding to the target text is extracted from the intermediate layer of the target self-encoding structure; wherein, the target self-encoding structure is used to Encoding and decoding the target text to obtain the hidden feature vector; the hidden feature vector is used to represent the semantic feature of the target text;
  • the similarity is taken as the matching degree between the target text and the target standard text.
  • FIG. 1 is an application environment diagram of a method for detecting text matching degree according to one or more embodiments
  • FIG. 2 is a schematic flowchart of a method for detecting text matching degree according to one or more embodiments
  • Fig. 3 is a schematic diagram of a self-encoding structure according to one or more embodiments.
  • FIG. 4 is a schematic flowchart of a method for detecting text matching degree in another embodiment
  • Fig. 5 is a block diagram of a text matching degree detection device according to one or more embodiments.
  • Figure 6 is a block diagram of a computer device according to one or more embodiments.
  • the text matching degree detection method provided in this application can be applied to the application environment as shown in FIG. 1.
  • the computer device may be a server, and its internal structure diagram may be as shown in Figure 1.
  • the computer equipment includes a processor, a memory, a network interface and a database connected through a system bus. Among them, the processor of the computer device is used to provide calculation and control capabilities.
  • the memory of the computer device includes a non-volatile computer-readable storage medium and internal memory.
  • the non-volatile computer-readable storage medium stores an operating system, computer-readable instructions, and a database.
  • the internal memory provides an environment for the operation of the operating system and computer-readable instructions in the non-volatile computer-readable storage medium.
  • the database of the computer equipment is used to store data such as voice and text.
  • the network interface of the computer device is used to communicate with an external terminal through a network connection.
  • the computer-readable instruction is executed by the processor to realize a text matching degree detection method.
  • the server obtains the input voice text and converts the voice text into the target text; the server inputs the target text into the preset target self-encoding structure, and extracts the hidden feature vector corresponding to the target text from the middle layer of the target self-encoding structure ;
  • the target self-encoding structure is used to encode and decode the target text to obtain the hidden feature vector;
  • the hidden feature vector is used to represent the semantic features of the target text;
  • the server obtains the preset target standard text corresponding to the target text ,
  • the target standard text is converted into the target standard vector; the server obtains the similarity between the hidden feature vector and the target standard vector, according to the correspondence between the hidden feature vector and the target text, and the difference between the target standard vector and the target standard text
  • the corresponding relationship between the two, the similarity is regarded as the matching degree between the
  • FIG. 1 is only a block diagram of part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied.
  • the specific computer device may Including more or fewer parts than shown in the figure, or combining some parts, or having a different arrangement of parts.
  • a method for detecting text matching degree is provided. Taking the method applied to the server in FIG. 1 as an example for description, the method includes the following steps:
  • Step S202 Obtain the input voice text, and convert the voice text into target text
  • Voice text refers to voice information
  • the target text is used to analyze the matching degree between the corresponding target standard text, and specifically refers to text information.
  • a voice conversion tool may be used to convert the voice text into the target text.
  • Step S204 input the target text into the preset target self-encoding structure, and extract the implicit feature vector corresponding to the target text from the middle layer of the target self-encoding structure; wherein, the target self-encoding structure is used to encode and decode the target text, To get the hidden feature vector; the hidden feature vector is used to represent the semantic features of the target text.
  • the self-encoding structure is a kind of neural network, as shown in Figure 3.
  • Figure 3 is a schematic diagram of the self-encoding structure. Input x 1 -x 6 into the self-encoding structure, and the self-encoding structure encodes x 1 -x 6 and then decodes , Get x 1 -x 6 , minimize the difference between the output x 1 -x 6 and the input x 1 -x 6 , then the middle layer of the self-encoding structure is the implicit of the input x 1 -x 6 Feature vector.
  • the implicit feature vector is the feature vector obtained by encoding the features of the input self-encoding structure, which retains a large amount of information of the input vector of the original input self-encoding structure, that is, the semantic features of the target text are retained; the self-encoding structure then includes the hidden features The vector is decoded and restored to obtain the output feature code.
  • the target text is converted into the target input code, and the target input code is input into the target self-encoding structure to obtain the implicit feature vector of the target text.
  • Step S206 Obtain a preset target standard text corresponding to the target text, and convert the target standard text into a target standard vector;
  • the target text is used to analyze the similarity between the corresponding target standard text.
  • Step S208 Obtain the similarity between the hidden feature vector and the target standard vector, and take the similarity as the corresponding relationship between the hidden feature vector and the target text, and the corresponding relationship between the target standard vector and the target standard text The degree of match between the target text and the target standard text.
  • the calculation of similarity is usually based on vectors. The distance between two vectors is calculated. The closer the distance, the greater the similarity.
  • the cosine similarity calculation method is used to calculate the similarity between the implicit feature vector and the target standard vector; according to the correspondence between the implicit feature vector and the target text, the target corresponding to the implicit feature vector is obtained Text; According to the correspondence between the target standard vector and the target standard text, the target standard text corresponding to the target standard vector is obtained; the target text corresponding to the hidden feature vector is based on the similarity between the hidden feature vector and the target standard vector, The target standard text corresponding to the target standard vector uses the similarity between the implicit feature vector and the target standard vector as the matching degree between the target text and the target standard text.
  • the cosine similarity calculation method is also called cosine similarity, which evaluates the similarity of two vectors by calculating the cosine of the angle between them.
  • the above-mentioned text matching degree detection method obtains the input speech text and converts the speech text into the target text; inputs the target text into the preset target self-encoding structure, and extracts the hidden meaning corresponding to the target text from the middle layer of the target self-encoding structure Contains feature vectors; among them, the target self-encoding structure is used to encode and decode the target text to obtain the hidden feature vector; the hidden feature vector is used to represent the semantic features of the target text; to obtain the preset target corresponding to the target text Standard text, convert the target standard text into a target standard vector; obtain the similarity between the hidden feature vector and the target standard vector, according to the correspondence between the hidden feature vector and the target text, as well as the target standard vector and the target standard text The correspondence relationship between the similarity is regarded as the matching degree between the target text and the target standard text.
  • the hidden features of the target text are extracted through the target self-encoding structure to obtain the semantic features of the target text, and then the similarity between the hidden feature vector of the target text and the target standard vector is calculated as the difference between the target text and the target labeled text
  • the matching degree can effectively retain the semantic features of the text and improve the accuracy of the matching degree detection results.
  • Fig. 4 is a schematic flowchart of a method for detecting text matching degree in another embodiment.
  • the target text is input into the preset target self-encoding structure, and before the hidden feature vector corresponding to the target text is extracted from the intermediate layer of the target self-encoding structure, include:
  • Step S203 Train a plurality of pre-stored self-encoding structures respectively, and obtain a target self-encoding structure from the self-encoding structure after training.
  • the step S203 of separately training a plurality of pre-stored auto-encoding structures, and obtaining the target auto-encoding structure from the trained auto-encoding structure includes:
  • the sample feature vector is obtained from the training text used to train the self-encoding structure.
  • the training text for the self-encoding structure after training is obtained, and the training text is converted into a sample feature vector through a preset learning algorithm.
  • a preset learning algorithm For example, through the sklearn library in Python, the target text is converted into the target input vector in the form of the feature of the bag of words model.
  • Python is a computer programming language
  • sklearn also known as scikit-learn
  • scikit-learn is a python-based machine learning library that can facilitate the implementation of machine learning algorithms, including: classification, regression, clustering, dimensionality reduction, model selection and Data mining related algorithms such as preprocessing.
  • the training process of the self-encoding structure is the process of minimizing the difference between input and output.
  • the sample feature vector is input into multiple different self-encoding structures.
  • the difference between the different self-encoding structures lies in the number of hidden layers and the number of hidden layer units. Adjust the parameters of multiple self-encoding structures to make The difference between the output of the respective coding structure and the sample feature vector is minimized.
  • the target auto-encoding structure is selected from multiple self-encoding structures after training.
  • the degree of difference between the input and output that can be achieved is also different. According to the input and output of each self-encoding structure after training The output difference value selects the target auto-encoding structure from multiple training auto-encoding structures.
  • inputting the target text into the preset target self-encoding structure in step S204, and extracting the implicit feature vector corresponding to the target text from the middle layer of the target self-encoding structure includes:
  • the preset learning algorithm is an algorithm used to convert text into a corresponding vector. For example, through the sklearn library in Python, the target text is converted into a target input vector in the form of a bag of words model feature.
  • Python is a computer programming language
  • sklearn also known as scikit-learn
  • scikit-learn is a python-based machine learning library that can facilitate the implementation of machine learning algorithms, including: classification, regression, clustering, dimensionality reduction, model selection and Data mining related algorithms such as preprocessing.
  • the existing text 1 “I like to eat apples, apples are rich in nutrition”
  • the text 2 “I like to eat pears”
  • the sklearn library uses the sklearn library to establish the features of the bag of words model (features will include “I”, “like”, “eat”, “apple”, “nutrition”, “rich”, and “pear"), and determine each sample according to the frequency of word occurrence
  • the feature value of can be obtained, the feature vector of text one (1,1,1,2,1,1,0), the feature vector of text two is (1,1,1,0,0,0,1)) .
  • the jieba library is a Python Chinese word segmentation library.
  • the hidden feature vector is the feature vector obtained by encoding the features of the input self-encoding structure, which retains a large amount of information of the input vector of the original input self-encoding structure; the self-encoding structure then decodes and restores the implicit feature vector to obtain the output feature code .
  • Figure 3 is a schematic diagram of the self-encoding structure.
  • the left side is the input
  • the right side is the output
  • the middle is the hidden feature vector.
  • Input the target input vector into the target encoding structure, and extract the implicitness of the middle layer of the target self-encoding structure Feature vector.
  • training multiple pre-stored auto-encoding structures based on the sample feature vector to obtain multiple self-encoding structures after training includes:
  • the server prestores a variety of self-encoding structures with different hidden layer structures, and respectively inputs the sample feature vectors into the self-encoding structures of multiple different hidden layers to obtain corresponding sample output vectors.
  • three self-encoding structures with one, two and three hidden layers are pre-stored, and the sample feature vectors are input into the three self-encoding structures to obtain the corresponding three sample output vectors.
  • the amount of information loss represents the information loss in the process of inputting the sample feature vector into the self-encoding structure, and then outputting the sample output vector.
  • the training process of the self-encoding structure is a process to minimize the difference between input and output
  • the information loss of the self-encoding structure should be calculated by the sample feature vector and the sample output vector.
  • the parameters are continuously adjusted to reduce the information loss until the information loss converges, that is, the reduction stops, indicating that the current parameters are the optimal parameters, and the training of the self-encoding structure is completed.
  • obtaining the amount of information loss between the sample feature vector and the sample output vector includes: using a mean square error loss function to calculate the amount of information loss between the sample feature vector and the sample output vector.
  • m is the number of samples
  • y a is the sample feature vector
  • y b is the sample output vector
  • the step of selecting a target auto-encoding structure from multiple trained auto-encoding structures according to the information loss rate of the trained auto-encoding structure includes:
  • the loss function of each auto-encoding model after training is still non-zero, and the trained auto-encoding structure with the lowest loss function is selected , That is, the self-encoding structure after training with the smallest difference between input and output, as the target self-encoding structure used.
  • calculating the similarity between the implicit feature vector and the target standard vector includes:
  • the cosine similarity calculation method is used to calculate the cosine value of the angle between the implicit feature vector and the target standard vector as the similarity between the implicit feature vector and the target standard vector.
  • the cosine similarity calculation method is also called cosine similarity. It evaluates the similarity of two vectors by calculating the cosine of the angle between them; the cosine of a 0 degree angle is 1, and the cosine of any other angle is not greater than 1. , And its minimum value is -1, so the cosine of the angle between the two vectors determines whether the two vectors are roughly pointing in the same direction.
  • the value of cosine similarity is 1; when the angle between the implicit feature vector and the target standard vector is 90°, the value of cosine similarity is 0; the two vectors point to perfect In the opposite direction, the value of cosine similarity is -1; cosine similarity is usually used in positive space, so the value given is between 0 and 1.
  • a text matching degree detection device which includes: a text acquisition module 502, an implicit feature acquisition module 504, a standard vector acquisition module 506, and a matching degree detection module 508, wherein :
  • the text obtaining module 502 is used to obtain the input voice text and convert the voice text into target text;
  • the hidden feature acquisition module 504 is used to input the target text into the preset target self-encoding structure, and extract the hidden feature vector corresponding to the target text from the middle layer of the target self-encoding structure; wherein the target self-encoding structure is used to The text is encoded and decoded to obtain the hidden feature vector; the hidden feature vector is used to represent the semantic features of the target text;
  • the standard vector obtaining module 506 is configured to obtain a preset target standard text corresponding to the target text, and convert the target standard text into a target standard vector;
  • the matching degree detection module 508 is used to obtain the similarity between the hidden feature vector and the target standard vector, according to the corresponding relationship between the hidden feature vector and the target text, and the corresponding relationship between the target standard vector and the target standard text , Regard the similarity as the matching degree between the target text and the target standard text.
  • the text matching degree detection device further includes a target self-encoding structure acquisition module, which is used for the implicit feature acquisition module to input the target text into the preset target self-encoding structure, and extract the data from the middle layer of the target self-encoding structure.
  • a target self-encoding structure acquisition module which is used for the implicit feature acquisition module to input the target text into the preset target self-encoding structure, and extract the data from the middle layer of the target self-encoding structure.
  • the target self-encoding structure acquisition module is also used to input multiple pre-stored sample feature vectors into the pre-stored self-encoding structure to obtain the corresponding sample output vector; according to the sample feature vector, the multiple pre-stored self-encoding The structure is trained to obtain multiple self-encoding structures after training; the target self-encoding structure is selected from the multiple self-encoding structures after training according to the information loss rate of the self-encoding structure after training.
  • the hidden feature acquisition module is also used to input the target text into the preset learning algorithm and output the target input vector; input the target input vector into the target self-encoding structure, and extract the target text from the middle layer of the target self-encoding structure The implicit feature vector corresponding to the input vector.
  • the target self-encoding structure acquisition module is also used to input the sample feature vector into the pre-stored self-encoding structure to obtain the corresponding sample output vector; acquire the information loss amount between the sample feature vector and the sample output vector; The amount of information loss adjusts the parameters of the self-encoding structure until the information loss amount obtained according to the self-encoding structure after adjusting the parameters meets the preset condition, and the self-encoding structure after the adjusted parameters is used as the self-encoding structure after training.
  • the target self-encoding structure acquisition module is also used to use the mean square error loss function to calculate the amount of information loss between the sample feature vector and the sample output vector; wherein the mean square error loss function formula is:
  • m is the number of samples
  • y a is the sample feature vector
  • y b is the sample output vector
  • the target self-encoding structure acquisition module is also used to compare the information loss of multiple trained self-encoding structures; select the trained self-encoding structure with the smallest amount of information loss as the target self-encoding structure .
  • Each module in the above-mentioned text matching degree detection device can be implemented in whole or in part by software, hardware and a combination thereof.
  • the foregoing modules may be embedded in the form of hardware or independent of the processor in the computer device, or may be stored in the memory of the computer device in the form of software, so that the processor can call and execute the operations corresponding to the foregoing modules.
  • a computer device is provided.
  • the computer device may be a server, and its internal structure diagram may be as shown in FIG. 6.
  • the computer equipment includes a processor, a memory, a network interface and a database connected by a device bus.
  • the processor of the computer device is used to provide calculation and control capabilities.
  • the memory of the computer device includes a non-volatile storage medium and an internal memory.
  • the non-volatile storage medium stores an operating device, computer-readable instructions, and a database.
  • the internal memory provides an environment for the operation of the operating device and computer-readable instructions in the non-volatile storage medium.
  • the database of the computer device is used to store data involved in text matching degree detection.
  • the network interface of the computer device is used to communicate with an external terminal through a network connection.
  • the computer readable instruction is executed by the processor to realize a text matching degree detection method.
  • FIG. 6 is only a block diagram of part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied.
  • the specific computer device may Including more or fewer parts than shown in the figure, or combining some parts, or having a different arrangement of parts.
  • a computer device includes a memory and one or more processors.
  • the memory stores computer readable instructions.
  • the one or more processors execute the following steps:
  • the target text is input into the preset target self-encoding structure, and the hidden feature vector corresponding to the target text is extracted from the middle layer of the target self-encoding structure; the target self-encoding structure is used to encode and decode the target text to obtain the hidden feature vector Contains feature vectors; implicit feature vectors are used to represent the semantic features of the target text;
  • the similarity is regarded as the target text and The degree of match between the target standard text.
  • the processor further implements the following steps when executing the computer-readable instructions: input the target text into the preset target self-encoding structure, and extract the hidden features corresponding to the target text from the middle layer of the target self-encoding structure Before vectoring, train multiple pre-stored self-encoding structures separately, and obtain the target self-encoding structure from the self-encoding structure after training.
  • the processor further implements the following steps when executing the computer-readable instructions: input a plurality of pre-stored sample feature vectors into the pre-stored self-encoding structure to obtain the corresponding sample output vector; according to the sample feature vector, the pre-stored Multiple self-encoding structures are trained to obtain multiple self-encoding structures after training; and the target self-encoding structure is selected from the multiple self-encoding structures after training according to the information loss rate of the self-encoding structure after training.
  • the processor further implements the following steps when executing the computer-readable instructions: input the target text into the preset learning algorithm, and output the target input vector; and input the target input vector into the target self-encoding structure, from the target self-encoding structure
  • the middle layer extracts the hidden feature vector corresponding to the target input vector.
  • the processor further implements the following steps when executing the computer-readable instructions: input the sample feature vector into the pre-stored self-encoding structure to obtain the corresponding sample output vector; obtain the information between the sample feature vector and the sample output vector The amount of loss; and adjust the parameters of the self-encoding structure according to the amount of information loss, until the amount of information loss obtained according to the self-encoding structure after adjusting the parameters meets the preset conditions, and the self-encoding structure after adjusting the parameters is used as the self-encoding structure after training.
  • the processor further implements the following steps when executing the computer-readable instructions: using the mean square error loss function to calculate the amount of information loss between the sample feature vector and the sample output vector; wherein, the mean square error loss function formula for:
  • m is the number of samples
  • y a is the sample feature vector
  • y b is the sample output vector
  • the processor further implements the following steps when executing the computer-readable instructions: comparing the information loss of multiple trained self-encoding structures; and selecting the trained self-encoding structure with the smallest amount of information loss, As the target self-encoding structure.
  • One or more non-volatile computer-readable storage media storing computer-readable instructions.
  • the one or more processors execute the following steps:
  • the target text is input into the preset target self-encoding structure, and the hidden feature vector corresponding to the target text is extracted from the middle layer of the target self-encoding structure; the target self-encoding structure is used to encode and decode the target text to obtain the hidden feature vector Contains feature vectors; implicit feature vectors are used to represent the semantic features of the target text;
  • the similarity is regarded as the target text and The degree of match between the target standard text.
  • the following steps are also implemented: when the target text is input into the preset target self-encoding structure, the implicit meaning corresponding to the target text is extracted from the middle layer of the target self-encoding structure Before the feature vector, a plurality of pre-stored auto-encoding structures are trained separately, and the target auto-encoding structure is obtained from the trained auto-encoding structure.
  • the following steps are also implemented: input multiple pre-stored sample feature vectors into the pre-stored self-encoding structure to obtain the corresponding sample output vector; according to the sample feature vector, the pre-stored The multiple self-encoding structures of the training are trained to obtain multiple self-encoding structures after training; and the target self-encoding structure is selected from the multiple trained self-encoding structures according to the information loss rate of the trained self-encoding structure.
  • the following steps are also implemented: input the target text into the preset learning algorithm, and output the target input vector; and input the target input vector into the target self-encoding structure, and then self-encode the target text.
  • the middle layer of the structure extracts the hidden feature vector corresponding to the target input vector.
  • the following steps are also implemented: input the sample feature vector into a pre-stored self-encoding structure to obtain the corresponding sample output vector; obtain the difference between the sample feature vector and the sample output vector The amount of information loss; and adjust the parameters of the self-encoding structure according to the amount of information loss until the amount of information loss obtained according to the self-encoding structure after adjusting the parameters meets the preset conditions, and the self-encoding structure after adjusting the parameters is used as the self-encoding structure after training .
  • the mean square error loss function is used to calculate the amount of information loss between the sample feature vector and the sample output vector; wherein, the mean square error loss function The formula is:
  • m is the number of samples
  • y a is the sample feature vector
  • y b is the sample output vector
  • the following steps are also implemented: comparing the information loss of multiple trained self-encoding structures; and selecting the trained self-encoding structure with the smallest amount of information loss , As the target self-encoding structure.
  • Non-volatile memory may include read only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory.
  • ROM read only memory
  • PROM programmable ROM
  • EPROM electrically programmable ROM
  • EEPROM electrically erasable programmable ROM
  • Volatile memory may include random access memory (RAM) or external cache memory.
  • RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous chain Channel (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Computational Linguistics (AREA)
  • Data Mining & Analysis (AREA)
  • General Engineering & Computer Science (AREA)
  • General Health & Medical Sciences (AREA)
  • Health & Medical Sciences (AREA)
  • Artificial Intelligence (AREA)
  • Biomedical Technology (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • Evolutionary Computation (AREA)
  • Biophysics (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Databases & Information Systems (AREA)
  • Acoustics & Sound (AREA)
  • Audiology, Speech & Language Pathology (AREA)
  • Compression, Expansion, Code Conversion, And Decoders (AREA)
  • Information Retrieval, Db Structures And Fs Structures Therefor (AREA)

Abstract

一种文本匹配度检测方法,包括:获取输入的语音文本,并将语音文本转为目标文本(S202);将目标文本输入预设的目标自编码结构,从目标自编码结构的中间层提取与目标文本对应的隐含特征向量;目标自编码结构用于对目标文本进行编码并解码,以得到隐含特征向量;隐含特征向量用于表示目标文本的语义特征(S204);获取与目标文本对应的预设的目标标准文本,将目标标准文本转化为目标标准向量(S206);获取隐含特征向量与目标标准向量之间的相似度,根据隐含特征向量和目标文本之间的对应关系,以及目标标准向量和目标标准文本之间的对应关系,将相似度作为目标文本和目标标准文本之间的匹配度(S208)。

Description

文本匹配度检测方法、装置、计算机设备和可读存储介质
本申请要求于2019年7月4日提交中国专利局,申请号为201910599137.6,申请名称为“文本匹配度检测方法、装置、计算机设备和可读存储介质”的中国专利申请的优先权,其全部内容通过引用结合在本申请中。
技术领域
本申请涉及一种文本匹配度检测方法、装置、计算机设备和可读存储介质。
背景技术
文本匹配度是指不同文本之间的语义关联度,文本匹配度的确定是文本挖掘和文本检索的核心工作之一;例如,利用语音关键字从数据库中检索相关文献时,就需要利用文本匹配度检测和语音关键字对应的文献。因此,如何更好的进行文本匹配度检测一直是本领域技术人员极为关注的问题。
现有技术进行文本匹配度检测的主要方式为:将文本映射成单词空间中的一个向量,计算向量之间的欧式距离或者余弦距离。但是,发明人意识到,现有的这种检测方式不能有效地提取文本的语义特征,导致文本匹配度的检测准确率比较低,从而造成匹配度检测结果的准确率较低。
发明内容
根据本申请公开的各种实施例,提供一种文本匹配度检测方法、装置、计算机设备和可读存储介质。
一种文本匹配度检测方法包括:
获取输入的语音文本,并将所述语音文本转为目标文本;
将所述目标文本输入预设的目标自编码结构,从所述目标自编码结构的中间层提取与所述目标文本对应的隐含特征向量;其中,所述目标自编码结构用于对所述目标文本进行编码并解码,以得到所述隐含特征向量;所述隐含特征向量用于表示所述目标文本的语义特征;
获取与所述目标文本对应的预设的目标标准文本,将所述目标标准文本转化为目标标准向量;及
获取所述隐含特征向量与所述目标标准向量之间的相似度,根据所述隐含特征向量和所述目标文本之间的对应关系,以及所述目标标准向量和所述目标标准文本之间的对应关系,将所述相似度作为所述目标文本和所述目标标准文本之间的匹配度。
一种文本匹配度检测装置包括:
文本获取模块,获取输入的语音文本,并将所述语音文本转为目标文本;
隐含特征获取模块,用于将所述目标文本输入预设的目标自编码结构,从所述目标自编码结构的中间层提取与所述目标文本对应的隐含特征向量;其中,所述目标自编码结构用于对所述目标文本进行编码并解码,以得到所述隐含特征向量;所述隐含特征向量用于表示所述目标文本的语义特征;
标准向量获取模块,用于获取与所述目标文本对应的预设的目标标准文本,将所述目标标准文本转化为目标标准向量;及
匹配度检测模块,用于获取所述隐含特征向量与所述目标标准向量之间的相似度,根据所述隐含特征向量和所述目标文本之间的对应关系,以及所述目标标准向量和所述目标标准文本之间的对应关系,将所述相似度作为所述目标文本和所述目标标准文本之间的匹配度。
一种计算机设备,包括存储器和一个或多个处理器,所述存储器中储存有计算机可读指令,所述计算机可读指令被所述处理器执行时,使得所述一个或多个处理器执行以下步骤:
获取输入的语音文本,并将所述语音文本转为目标文本;
将所述目标文本输入预设的目标自编码结构,从所述目标自编码结构的中间层提取与所述目标文本对应的隐含特征向量;其中,所述目标自编码结构用于对所述目标文本进行编码并解码,以得到所述隐含特征向量;所述隐含特征向量用于表示所述目标文本的语义特征;
获取与所述目标文本对应的预设的目标标准文本,将所述目标标准文本转化为目标标准向量;及
获取所述隐含特征向量与所述目标标准向量之间的相似度,根据所述隐含特征向量和所述目标文本之间的对应关系,以及所述目标标准向量和所述目标标准文本之间的对应关系,将所述相似度作为所述目标文本和所述目标标准文本之间的匹配度。
一个或多个存储有计算机可读指令的非易失性计算机可读存储介质,计算机可读指令被一个或多个处理器执行时,使得一个或多个处理器执行以下步骤:
获取输入的语音文本,并将所述语音文本转为目标文本;
将所述目标文本输入预设的目标自编码结构,从所述目标自编码结构的中间层提取与所述目标文本对应的隐含特征向量;其中,所述目标自编码结构用于对所述目标文本进行编码并解码,以得到所述隐含特征向量;所述隐含特征向量用于表示所述目标文本的语义特征;
获取与所述目标文本对应的预设的目标标准文本,将所述目标标准文本转化为目标标准向量;及
获取所述隐含特征向量与所述目标标准向量之间的相似度,根据所述隐含特征向量和所述目标文本之间的对应关系,以及所述目标标准向量和所述目标标准文本之间的对应关系,将所述相似度作为所述目标文本和所述目标标准文本之间的匹配度。
本申请的一个或多个实施例的细节在下面的附图和描述中提出。本申请的其它特征和优点将从说明书、附图以及权利要求书变得明显。
附图说明
为了更清楚地说明本申请实施例中的技术方案,下面将对实施例中所需要使用的附图作简单地介绍,显而易见地,下面描述中的附图仅仅是本申请的一些实施例,对于本领域普通技术人员来讲,在不付出创造性劳动的前提下,还可以根据这些附图获得其它的附图。
图1为根据一个或多个实施例中文本匹配度检测方法的应用环境图;
图2为根据一个或多个实施例中文本匹配度检测方法的流程示意图;
图3为根据一个或多个实施例中自编码结构的示意图;
图4为另一个实施例中文本匹配度检测方法的流程示意图;
图5为根据一个或多个实施例中文本匹配度检测装置的框图;
图6为根据一个或多个实施例中计算机设备的框图。
具体实施方式
为了使本申请的技术方案及优点更加清楚明白,以下结合附图及实施例,对本申请进行进一步详细说明。应当理解,此处描述的具体实施例仅仅用以解释本申请,并不用于限定本申请。
本申请提供的文本匹配度检测方法,可以应用于如图1所示的应用环境中。该计算机设备可以是服务器,其内部结构图可以如图1所示。该计算机设备包括通过系统总线连接的处理器、存储器、网络接口和数据库。其中,该计算机设备的处理器用于提供计算和控制能力。该计算机设备的存储器包括非易失性计算机可读存储介质、内存储器。该非易失性计算机可读存储介质存储有操作系统、计算机可读指令和数据库。该内存储器为非易失性计算机可读存储介质中的操作系统和计算机可读指令的运行提供环境。该计算机设备的数据库用于存储语音文本等数据。该计算机设备的网络接口用于与外部的终端通过网络连接通信。该计算机可读指令被处理器执行时以实现一种文本匹配度检测方法。具体地,服务器获取输入的语音文本,并将语音文本转为目标文本;服务器将目标文本输入预设的目标自编码结构,从目标自编码结构的中间层提取与目标文本对应的隐含特征向量;其中,目标自编码结构用于对目标文本进行编码并解码,以得到隐含特征向量;隐含特征向量用于表示目标文本的语义特征;服务器获取与目标文本对应的预设的目标标准文本,将目标标准文本转化为目标标准向量;服务器获取隐含特征向量与目标标准向量之间的相似度,根据隐含特征向量和目标文本之间的对应关系,以及目标标准向量和目标标准文本之间的对应关系,将相似度作为目标文本和目标标准文本之间的匹配度。服务器可以用独立的服务器或者是多个服务器组成的服务器集群来实现。
本领域技术人员可以理解,图1中示出的结构,仅仅是与本申请方案相关的部分结构 的框图,并不构成对本申请方案所应用于其上的计算机设备的限定,具体的计算机设备可以包括比图中所示更多或更少的部件,或者组合某些部件,或者具有不同的部件布置。
在其中一个实施例中,如图2所示,提供了一种文本匹配度检测方法,以该方法应用于图1中的服务器为例进行说明,包括以下步骤:
步骤S202,获取输入的语音文本,并将语音文本转为目标文本;
语音文本是指语音信息,目标文本用于分析和对应的目标标准文本之间的匹配度,具体是指文字信息。
具体地,可以采用语音转化工具将语音文本转化为目标文本。
步骤S204,将目标文本输入预设的目标自编码结构,从目标自编码结构的中间层提取与目标文本对应的隐含特征向量;其中,目标自编码结构用于对目标文本进行编码并解码,以得到隐含特征向量;隐含特征向量用于表示目标文本的语义特征。
自编码结构是一种神经网络,如图3所示,图3为自编码结构的示意图,将x 1-x 6输入自编码结构,自编码结构对x 1-x 6进行编码,然后进行解码,得到x 1-x 6,使得输出的x 1-x 6和输入的x 1-x 6之间的差异最小化,此时自编码结构的中间层就是输入的x 1-x 6的隐含特征向量。
隐含特征向量是将输入自编码结构的特征进行编码得到的特征向量,保留了原始输入自编码结构的输入向量的大量信息,即保留了目标文本的语义特征;自编码结构再将隐含特征向量进行解码还原,得到输出特征编码。
在具体实施过程中,通过将目标文本转化为目标输入编码,将目标输入编码输入目标自编码结构,得到目标文本的隐含特征向量。
步骤S206,获取与目标文本对应的预设的目标标准文本,将目标标准文本转化为目标标准向量;
目标文本用于分析和对应的目标标准文本之间的相似度。
在具体实施过程中,由于文本本身不便于直接进行相似度对比,因此,需要将预设的目标标准文本转换成目标标注向量,然后和隐含特征向量进行对比获取相似度。
步骤S208,获取隐含特征向量与目标标准向量之间的相似度,根据隐含特征向量和目标文本之间的对应关系,以及目标标准向量和目标标准文本之间的对应关系,将相似度作为目标文本和目标标准文本之间的匹配度。
关于相似度的计算,通常是基于向量,计算两个向量之间的距离,距离越近,相似度越大。
在具体实施过程中,采用余弦相似度计算方法,计算隐含特征向量和目标标准向量之间的相似度;根据隐含特征向量和目标文本之间的对应关系,得到隐含特征向量对应的目标文本;根据目标标准向量和目标标准文本之间的对应关系,得到目标标准向量对应的目标标准文本;基于隐含特征向量和目标标准向量之间的相似度、隐含特征向量对应的目标文本、目标标准向量对应的目标标准文本,将隐含特征向量和目标标准向量之间的相似度, 作为目标文本和目标标准文本之间的匹配度。
余弦相似度计算方法又称为余弦相似性,是通过计算两个向量的夹角余弦值来评估他们的相似度。
上述文本匹配度检测方法,通过获取输入的语音文本,并将语音文本转为目标文本;将目标文本输入预设的目标自编码结构,从目标自编码结构的中间层提取与目标文本对应的隐含特征向量;其中,目标自编码结构用于对目标文本进行编码并解码,以得到隐含特征向量;隐含特征向量用于表示目标文本的语义特征;获取与目标文本对应的预设的目标标准文本,将目标标准文本转化为目标标准向量;获取隐含特征向量与目标标准向量之间的相似度,根据隐含特征向量和目标文本之间的对应关系,以及目标标准向量和目标标准文本之间的对应关系,将相似度作为目标文本和目标标准文本之间的匹配度。通过目标自编码结构提取目标文本的隐含特征,从而得到目标文本的语义特征,然后计算目标文本的隐含特征向量和目标标准向量之间的相似度,作为目标文本和目标标注文本之间的匹配度,可以有效保留文本的语义特征,提高匹配度检测结果准确率。
图4为另一个实施例中文本匹配度检测方法的流程示意图。如图4所示,在其中一个实施例中,步骤S204的将目标文本输入预设的目标自编码结构,从目标自编码结构的中间层提取与目标文本对应的隐含特征向量的之前,还包括:
步骤S203,对多个预存的自编码结构分别进行训练,并从训练后的自编码结构中获取目标自编码结构。
具体地,步骤S203对多个预存的自编码结构分别进行训练,并从训练后的自编码结构中获取目标自编码结构的步骤,包括:
(1)将预存的多个样本特征向量输入预存的自编码结构,得到对应的样本输出向量。
样本特征向量,是根据用于对自编码结构进行训练的训练文本得到的。
在具体实施过程中,获取用于训练后的自编码结构的训练文本,通过预设学习算法将训练文本转化成样本特征向量。例如,通过Python中的sklearn库,将目标文本转化为词袋模型特征形式的目标输入向量。
Python是一种计算机程序设计语言;sklearn,也称为scikit-learn,是基于python的机器学习库,可以方便进行机器学习算法的实施,包括:分类、回归、聚类、降维、模型选择和预处理等数据挖掘的相关算法。
(2)根据样本特征向量,对预存的多个自编码结构进行训练,得到多个训练后的自编码结构。
自编码结构的训练过程是使输入和输出差异最小化的过程。
在具体实施过程中,将样本特征向量分别输入多个不同的自编码结构,不同的自编码结构的区别在于隐层数量和隐层单元数量的不同,分别调整多个自编码结构的参数,使各自编码结构输出和样本特征向量差异最小化。
(3)根据训练后的自编码结构的信息损失率从多个训练后的自编码结构中选取目标 自编码结构。
在具体实施过程中,由于多个训练后的自编码结构的隐层和隐层单元数据的不同,因此,能达到的输入和输出差异程度也不同,根据各训练后的自编码结构的输入和输出的差异值,从多个训练后的自编码结构中选取目标自编码结构。
在其中一个实施例中,步骤S204的将目标文本输入预设的目标自编码结构,从目标自编码结构的中间层提取与目标文本对应的隐含特征向量,包括:
(1)将目标文本输入预设学习算法,输出目标输入向量。
预设学习算法是用于将文本转化成对应的向量的算法,例如,通过Python中的sklearn库,将目标文本转化为词袋模型特征形式的目标输入向量。
Python是一种计算机程序设计语言;sklearn,也称为scikit-learn,是基于python的机器学习库,可以方便进行机器学习算法的实施,包括:分类、回归、聚类、降维、模型选择和预处理等数据挖掘的相关算法。
例如,现有文本一:“我喜欢吃苹果,苹果营养丰富”,以及文本二:“我喜欢吃梨”,则首先通过Python中的jieba库进行分词以将话语中的词语分隔开来,随后通过sklearn库建立词袋模型特征(特征将包括“我”、“喜欢”、“吃”、“苹果”、“营养”、“丰富”、“梨”),并根据词语出现频次确定各个样本的特征数值,可以得到,文本一的特征向量(1,1,1,2,1,1,0),文本二的特征向量为(1,1,1,0,0,0,1))。其中,jieba库是一种Python中文分词库。
(2)将目标输入向量输入目标自编码结构,从目标自编码结构的中间层提取与目标输入向量对应的隐含特征向量。
隐含特征向量是将输入自编码结构的特征进行编码得到的特征向量,保留了原始输入自编码结构的输入向量的大量信息;自编码结构再将隐含特征向量进行解码还原,得到输出特征编码。
如图3所示,图3为自编码结构的结构示意图,左边为输入,右边为输出,中间为隐含特征向量,将目标输入向量输入目标编码结构,提取目标自编码结构中间层的隐含特征向量。
在其中一个实施例中,根据样本特征向量,对预存的多个自编码结构进行训练,得到多个训练后的自编码结构,包括:
(1)将样本特征向量输入预存的自编码结构,得到对应的样本输出向量。
在具体实施过程中,服务器预存有多种不同隐层结构的自编码结构,分别将样本特征向量输入多个不同隐层的自编码结构,得到对应的样本输出向量。
例如,预存有隐层数量分别为一层、两层和三层的三种自编码结构,将样本特征向量分别输入三种自编码结构,得到对应的三种样本输出向量。
(2)获取样本特征向量和样本输出向量之间的信息损失量。
信息损失量代表样本特征向量输入自编码结构,然后输出样本输出向量的这个过程中的信息损失。
在具体实施过程中,由于自编码结构的训练过程是使输入和输出差异最小化的过程,因此,要通过样本特征向量和样本输出向量计算自编码结构的信息损失量。
(3)根据信息损失量调整自编码结构的参数,直至根据调整参数后的自编码结构得到的信息损失量符合预设条件,将调整参数后的自编码结构作为训练后的自编码结构。
在具体实施过程中,每次输入样本特征向量,得到样本输出向量,计算样本特征向量和样本输出向量之间的信息损失,调整参数,重新输入样本特征向量进行训练,再次计算样本特征向量和样本输出向量之间的信息损失,不断调整参数,使信息损失不断减小,直到信息损失收敛,即停止减小,说明当前的参数为最优参数,自编码结构的训练完成。
在其中一个实施例中,获取样本特征向量和样本输出向量之间的信息损失量,包括:采用均方误差损失函数,计算样本特征向量和样本输出向量之间的信息损失量。
均方误差损失函数公式为:
Figure PCTCN2019103474-appb-000001
式中,m为样本数,y a为样本特征向量,y b为样本输出向量。
在具体实施过程中,样本特征向量和样本输出向量之间的差异越小,即y a和y b越接近,对应的自编码结构的信息损失量越小。
在其中一个实施例中,根据训练后的自编码结构的信息损失率从多个训练后的自编码结构中选取目标自编码结构的步骤,包括:
将多个训练后的自编码结构的信息损失量进行比较;选取信息损失量最小的训练后的自编码结构,作为目标自编码结构。
在具体实施过程中,由于训练后的自编码模型的输出无法实现对于输入的完全还原,因此训练完毕的各个自编码模型的损失函数依旧为非零,选取损失函数最低的训练后的自编码结构,也就是输入和输出差异最小的训练后的自编码结构,作为所用的目标自编码结构。
在其中一个实施例中,计算隐含特征向量与目标标准向量之间的相似度,包括:
采用余弦相似度计算方法,计算隐含特征向量与目标标准向量的夹角余弦值,作为隐含特征向量与目标标准向量之间的相似度。
余弦相似度计算方法又称为余弦相似性,是通过计算两个向量的夹角余弦值来评估他们的相似度;0度角的余弦值是1,而其他任何角度的余弦值都不大于1,并且其最小值是-1,从而两个向量之间的角度的余弦值确定两个向量是否大致指向相同的方向。隐含特征向量与目标标准向量有相同的指向时,余弦相似度的值为1;隐含特征向量与目标标准向量夹角为90°时,余弦相似度的值为0;两个向量指向完全相反的方向时,余弦相似度的值为-1;余弦相似度通常用于正空间,因此给出的值为0到1之间。
应该理解的是,虽然图2和图4的流程图中的各个步骤按照箭头的指示依次显示,但是这些步骤并不是必然按照箭头指示的顺序依次执行。除非本文中有明确的说明,这些步 骤的执行并没有严格的顺序限制,这些步骤可以以其它的顺序执行。而且,图2和图4中的至少一部分步骤可以包括多个子步骤或者多个阶段,这些子步骤或者阶段并不必然是在同一时刻执行完成,而是可以在不同的时刻执行,这些子步骤或者阶段的执行顺序也不必然是依次进行,而是可以与其它步骤或者其它步骤的子步骤或者阶段的至少一部分轮流或者交替地执行。
在其中一个实施例中,如图5所示,提供了一种文本匹配度检测装置,包括:文本获取模块502、隐含特征获取模块504、标准向量获取模块506和匹配度检测模块508,其中:
文本获取模块502,用于获取输入的语音文本,并将语音文本转为目标文本;
隐含特征获取模块504,用于将目标文本输入预设的目标自编码结构,从目标自编码结构的中间层提取与目标文本对应的隐含特征向量;其中,目标自编码结构用于对目标文本进行编码并解码,以得到隐含特征向量;隐含特征向量用于表示目标文本的语义特征;
标准向量获取模块506,用于获取与目标文本对应的预设的目标标准文本,将目标标准文本转化为目标标准向量;
匹配度检测模块508,用于获取隐含特征向量与目标标准向量之间的相似度,根据隐含特征向量和目标文本之间的对应关系,以及目标标准向量和目标标准文本之间的对应关系,将相似度作为目标文本和目标标准文本之间的匹配度。
在其中一个实施例中,文本匹配度检测装置还包括目标自编码结构获取模块,用于隐含特征获取模块将目标文本输入预设的目标自编码结构,从目标自编码结构的中间层提取与目标文本对应的隐含特征向量之前,对多个预存的自编码结构分别进行训练,并从训练后的自编码结构中获取目标自编码结构。
在其中一个实施例中,目标自编码结构获取模块还用于将预存的多个样本特征向量输入预存的自编码结构,得到对应的样本输出向量;根据样本特征向量,对预存的多个自编码结构进行训练,得到多个训练后的自编码结构;根据训练后的自编码结构的信息损失率从多个训练后的自编码结构中选取目标自编码结构。
在其中一个实施例中,隐含特征获取模块还用于将目标文本输入预设学习算法,输出目标输入向量;将目标输入向量输入目标自编码结构,从目标自编码结构的中间层提取与目标输入向量对应的隐含特征向量。
在其中一个实施例中,目标自编码结构获取模块还用于将样本特征向量输入预存的自编码结构,得到对应的样本输出向量;获取样本特征向量和样本输出向量之间的信息损失量;根据信息损失量调整自编码结构的参数,直至根据调整参数后的自编码结构得到的信息损失量符合预设条件,将调整参数后的自编码结构作为训练后的自编码结构。
在其中一个实施例中,目标自编码结构获取模块还用于采用均方误差损失函数,计算样本特征向量和样本输出向量之间的信息损失量;其中,均方误差损失函数公式为:
Figure PCTCN2019103474-appb-000002
式中,m为样本数,y a为样本特征向量,y b为样本输出向量。
在其中一个实施例中,目标自编码结构获取模块还用于将多个训练后的自编码结构的信息损失量进行比较;选取信息损失量最小的训练后的自编码结构,作为目标自编码结构。
关于文本匹配度检测装置的具体限定可以参见上文中对于文本匹配度检测方法的限定,在此不再赘述。上述文本匹配度检测装置中的各个模块可全部或部分通过软件、硬件及其组合来实现。上述各模块可以硬件形式内嵌于或独立于计算机设备中的处理器中,也可以以软件形式存储于计算机设备中的存储器中,以便于处理器调用执行以上各个模块对应的操作。
在其中一个实施例中,提供了一种计算机设备,该计算机设备可以是服务器,其内部结构图可以如图6所示。该计算机设备包括通过装置总线连接的处理器、存储器、网络接口和数据库。该计算机设备的处理器用于提供计算和控制能力。该计算机设备的存储器包括非易失性存储介质、内存储器。该非易失性存储介质存储有操作装置、计算机可读指令和数据库。该内存储器为非易失性存储介质中的操作装置和计算机可读指令的运行提供环境。该计算机设备的数据库用于存储文本匹配度检测涉及的数据。该计算机设备的网络接口用于与外部的终端通过网络连接通信。该计算机可读指令被处理器执行时以实现一种文本匹配度检测方法。
本领域技术人员可以理解,图6中示出的结构,仅仅是与本申请方案相关的部分结构的框图,并不构成对本申请方案所应用于其上的计算机设备的限定,具体的计算机设备可以包括比图中所示更多或更少的部件,或者组合某些部件,或者具有不同的部件布置。
一种计算机设备,包括存储器和一个或多个处理器,存储器中储存有计算机可读指令,计算机可读指令被处理器执行时,使得一个或多个处理器执行以下步骤:
获取输入的语音文本,并将语音文本转为目标文本;
将目标文本输入预设的目标自编码结构,从目标自编码结构的中间层提取与目标文本对应的隐含特征向量;其中,目标自编码结构用于对目标文本进行编码并解码,以得到隐含特征向量;隐含特征向量用于表示目标文本的语义特征;
获取与目标文本对应的预设的目标标准文本,将目标标准文本转化为目标标准向量;及
获取隐含特征向量与目标标准向量之间的相似度,根据隐含特征向量和目标文本之间的对应关系,以及目标标准向量和目标标准文本之间的对应关系,将相似度作为目标文本和目标标准文本之间的匹配度。
在其中一个实施例中,处理器执行计算机可读指令时还实现以下步骤:在将目标文本输入预设的目标自编码结构,从目标自编码结构的中间层提取与目标文本对应的隐含特征向量之前,对多个预存的自编码结构分别进行训练,并从训练后的自编码结构中获取目标 自编码结构。
在其中一个实施例中,处理器执行计算机可读指令时还实现以下步骤:将预存的多个样本特征向量输入预存的自编码结构,得到对应的样本输出向量;根据样本特征向量,对预存的多个自编码结构进行训练,得到多个训练后的自编码结构;及根据训练后的自编码结构的信息损失率从多个训练后的自编码结构中选取目标自编码结构。
在其中一个实施例中,处理器执行计算机可读指令时还实现以下步骤:将目标文本输入预设学习算法,输出目标输入向量;及将目标输入向量输入目标自编码结构,从目标自编码结构的中间层提取与目标输入向量对应的隐含特征向量。
在其中一个实施例中,处理器执行计算机可读指令时还实现以下步骤:将样本特征向量输入预存的自编码结构,得到对应的样本输出向量;获取样本特征向量和样本输出向量之间的信息损失量;及根据信息损失量调整自编码结构的参数,直至根据调整参数后的自编码结构得到的信息损失量符合预设条件,将调整参数后的自编码结构作为训练后的自编码结构。
在其中一个实施例中,处理器执行计算机可读指令时还实现以下步骤:采用均方误差损失函数,计算样本特征向量和样本输出向量之间的信息损失量;其中,均方误差损失函数公式为:
Figure PCTCN2019103474-appb-000003
式中,m为样本数,y a为样本特征向量,y b为样本输出向量。
在其中一个实施例中,处理器执行计算机可读指令时还实现以下步骤:将多个训练后的自编码结构的信息损失量进行比较;及选取信息损失量最小的训练后的自编码结构,作为目标自编码结构。
一个或多个存储有计算机可读指令的非易失性计算机可读存储介质,计算机可读指令被一个或多个处理器执行时,使得一个或多个处理器执行以下步骤:
获取输入的语音文本,并将语音文本转为目标文本;
将目标文本输入预设的目标自编码结构,从目标自编码结构的中间层提取与目标文本对应的隐含特征向量;其中,目标自编码结构用于对目标文本进行编码并解码,以得到隐含特征向量;隐含特征向量用于表示目标文本的语义特征;
获取与目标文本对应的预设的目标标准文本,将目标标准文本转化为目标标准向量;及
获取隐含特征向量与目标标准向量之间的相似度,根据隐含特征向量和目标文本之间的对应关系,以及目标标准向量和目标标准文本之间的对应关系,将相似度作为目标文本和目标标准文本之间的匹配度。
在其中一个实施例中,计算机可读指令被处理器执行时还实现以下步骤:在将目标文本输入预设的目标自编码结构,从目标自编码结构的中间层提取与目标文本对应的隐含特 征向量之前,对多个预存的自编码结构分别进行训练,并从训练后的自编码结构中获取目标自编码结构。
在其中一个实施例中,计算机可读指令被处理器执行时还实现以下步骤:将预存的多个样本特征向量输入预存的自编码结构,得到对应的样本输出向量;根据样本特征向量,对预存的多个自编码结构进行训练,得到多个训练后的自编码结构;及根据训练后的自编码结构的信息损失率从多个训练后的自编码结构中选取目标自编码结构。
在其中一个实施例中,计算机可读指令被处理器执行时还实现以下步骤:将目标文本输入预设学习算法,输出目标输入向量;及将目标输入向量输入目标自编码结构,从目标自编码结构的中间层提取与目标输入向量对应的隐含特征向量。
在其中一个实施例中,计算机可读指令被处理器执行时还实现以下步骤:将样本特征向量输入预存的自编码结构,得到对应的样本输出向量;获取样本特征向量和样本输出向量之间的信息损失量;及根据信息损失量调整自编码结构的参数,直至根据调整参数后的自编码结构得到的信息损失量符合预设条件,将调整参数后的自编码结构作为训练后的自编码结构。
在其中一个实施例中,计算机可读指令被处理器执行时还实现以下步骤:采用均方误差损失函数,计算样本特征向量和样本输出向量之间的信息损失量;其中,均方误差损失函数公式为:
Figure PCTCN2019103474-appb-000004
式中,m为样本数,y a为样本特征向量,y b为样本输出向量。
在其中一个实施例中,计算机可读指令被处理器执行时还实现以下步骤:将多个训练后的自编码结构的信息损失量进行比较;及选取信息损失量最小的训练后的自编码结构,作为目标自编码结构。
本领域普通技术人员可以理解实现上述实施例方法中的全部或部分流程,是可以通过计算机可读指令来指令相关的硬件来完成,所述的计算机可读指令可存储于一非易失性计算机可读取存储介质中,该计算机可读指令在执行时,可包括如上述各方法的实施例的流程。其中,本申请所提供的各实施例中所使用的对存储器、存储、数据库或其它介质的任何引用,均可包括非易失性和/或易失性存储器。非易失性存储器可包括只读存储器(ROM)、可编程ROM(PROM)、电可编程ROM(EPROM)、电可擦除可编程ROM(EEPROM)或闪存。易失性存储器可包括随机存取存储器(RAM)或者外部高速缓冲存储器。作为说明而非局限,RAM以多种形式可得,诸如静态RAM(SRAM)、动态RAM(DRAM)、同步DRAM(SDRAM)、双数据率SDRAM(DDRSDRAM)、增强型SDRAM(ESDRAM)、同步链路(Synchlink)DRAM(SLDRAM)、存储器总线(Rambus)直接RAM(RDRAM)、直接存储器总线动态RAM(DRDRAM)、以及存储器总线动态RAM(RDRAM)等。
以上实施例的各技术特征可以进行任意的组合,为使描述简洁,未对上述实施例中的 各个技术特征所有可能的组合都进行描述,然而,只要这些技术特征的组合不存在矛盾,都应当认为是本说明书记载的范围。
以上所述实施例仅表达了本申请的几种实施方式,其描述较为具体和详细,但并不能因此而理解为对发明专利范围的限制。应当指出的是,对于本领域的普通技术人员来说,在不脱离本申请构思的前提下,还可以做出若干变形和改进,这些都属于本申请的保护范围。因此,本申请专利的保护范围应以所附权利要求为准。

Claims (20)

  1. 一种文本匹配度检测方法,包括:
    获取输入的语音文本,并将所述语音文本转为目标文本;
    将所述目标文本输入预设的目标自编码结构,从所述目标自编码结构的中间层提取与所述目标文本对应的隐含特征向量;其中,所述目标自编码结构用于对所述目标文本进行编码并解码,以得到所述隐含特征向量;所述隐含特征向量用于表示所述目标文本的语义特征;
    获取与所述目标文本对应的预设的目标标准文本,将所述目标标准文本转化为目标标准向量;及
    获取所述隐含特征向量与所述目标标准向量之间的相似度,根据所述隐含特征向量和所述目标文本之间的对应关系,以及所述目标标准向量和所述目标标准文本之间的对应关系,将所述相似度作为所述目标文本和所述目标标准文本之间的匹配度。
  2. 根据权利要求1所述的方法,其特征在于,在将所述目标文本输入预设的目标自编码结构,从所述目标自编码结构的中间层提取与所述目标文本对应的隐含特征向量之前,所述方法还包括:
    对多个预存的自编码结构分别进行训练,并从训练后的自编码结构中获取所述目标自编码结构。
  3. 根据权利要求2所述的方法,其特征在于,所述对多个预存的自编码结构分别进行训练,并从训练后的自编码结构中获取所述目标自编码结构,包括:
    将预存的多个样本特征向量输入预存的自编码结构,得到对应的样本输出向量;
    根据所述样本特征向量,对预存的多个自编码结构进行训练,得到多个训练后的自编码结构;及
    根据所述训练后的自编码结构的信息损失率从所述多个训练后的自编码结构中选取目标自编码结构。
  4. 根据权利要求1所述的方法,其特征在于,所述将所述目标文本输入预设的目标自编码结构,从所述目标自编码结构的中间层提取与所述目标文本对应的隐含特征向量,包括:
    将所述目标文本输入预设学习算法,输出目标输入向量;及
    将所述目标输入向量输入所述目标自编码结构,从所述目标自编码结构的中间层提取与所述目标输入向量对应的隐含特征向量。
  5. 根据权利要求3所述的方法,其特征在于,所述根据所述样本特征向量,对预存的多个自编码结构进行训练,得到多个训练后的自编码结构,包括:
    将所述样本特征向量输入预存的自编码结构,得到对应的样本输出向量;
    获取所述样本特征向量和所述样本输出向量之间的信息损失量;及
    根据所述信息损失量调整所述自编码结构的参数,直至根据调整参数后的自编码结构 得到的信息损失量符合预设条件,将所述调整参数后的自编码结构作为所述训练后的自编码结构。
  6. 根据权利要求5所述的方法,其特征在于,所述获取所述样本特征向量和所述样本输出向量之间的信息损失量,包括:
    采用均方误差损失函数,计算所述样本特征向量和所述样本输出向量之间的信息损失量;其中,所述均方误差损失函数公式为:
    Figure PCTCN2019103474-appb-100001
    式中,m为样本数,y a为样本特征向量,y b为样本输出向量。
  7. 根据权利要求3所述的方法,其特征在于,所述根据所述训练后的自编码结构的信息损失率从所述多个训练后的自编码结构中选取目标自编码结构,包括:
    将所述多个训练后的自编码结构的所述信息损失量进行比较;及
    选取信息损失量最小的训练后的自编码结构,作为所述目标自编码结构。
  8. 一种文本匹配度检测装置,包括:
    文本获取模块,获取输入的语音文本,并将所述语音文本转为目标文本;
    隐含特征获取模块,用于将所述目标文本输入预设的目标自编码结构,从所述目标自编码结构的中间层提取与所述目标文本对应的隐含特征向量;其中,所述目标自编码结构用于对所述目标文本进行编码并解码,以得到所述隐含特征向量;所述隐含特征向量用于表示所述目标文本的语义特征;
    标准向量获取模块,用于获取与所述目标文本对应的预设的目标标准文本,将所述目标标准文本转化为目标标准向量;及
    匹配度检测模块,用于获取所述隐含特征向量与所述目标标准向量之间的相似度,根据所述隐含特征向量和所述目标文本之间的对应关系,以及所述目标标准向量和所述目标标准文本之间的对应关系,将所述相似度作为所述目标文本和所述目标标准文本之间的匹配度。
  9. 根据权利要求8所述的装置,其特征在于,所述装置还包括目标自编码结构获取模块,用于隐含特征获取模块将目标文本输入预设的目标自编码结构,从目标自编码结构的中间层提取与目标文本对应的隐含特征向量之前,对多个预存的自编码结构分别进行训练,并从训练后的自编码结构中获取目标自编码结构。
  10. 根据权利要求9所述的装置,其特征在于,所述目标自编码结构获取模块还用于将预存的多个样本特征向量输入预存的自编码结构,得到对应的样本输出向量;根据所述样本特征向量,对预存的多个自编码结构进行训练,得到多个训练后的自编码结构;及根据所述训练后的自编码结构的信息损失率从所述多个训练后的自编码结构中选取目标自编码结构。
  11. 一种计算机设备,包括存储器及一个或多个处理器,所述存储器中储存有计算机 可读指令,所述计算机可读指令被所述一个或多个处理器执行时,使得所述一个或多个处理器执行以下步骤:
    获取输入的语音文本,并将所述语音文本转为目标文本;
    将所述目标文本输入预设的目标自编码结构,从所述目标自编码结构的中间层提取与所述目标文本对应的隐含特征向量;其中,所述目标自编码结构用于对所述目标文本进行编码并解码,以得到所述隐含特征向量;所述隐含特征向量用于表示所述目标文本的语义特征;
    获取与所述目标文本对应的预设的目标标准文本,将所述目标标准文本转化为目标标准向量;及
    获取所述隐含特征向量与所述目标标准向量之间的相似度,根据所述隐含特征向量和所述目标文本之间的对应关系,以及所述目标标准向量和所述目标标准文本之间的对应关系,将所述相似度作为所述目标文本和所述目标标准文本之间的匹配度。
  12. 根据权利要求11所述的计算机设备,其特征在于,所述处理器执行所述计算机可读指令时还执行以下步骤:在将所述目标文本输入预设的目标自编码结构,从所述目标自编码结构的中间层提取与所述目标文本对应的隐含特征向量之前,对多个预存的自编码结构分别进行训练,并从训练后的自编码结构中获取所述目标自编码结构。
  13. 根据权利要求12所述的计算机设备,其特征在于,所述处理器执行所述计算机可读指令时还执行以下步骤:
    将预存的多个样本特征向量输入预存的自编码结构,得到对应的样本输出向量;
    根据所述样本特征向量,对预存的多个自编码结构进行训练,得到多个训练后的自编码结构;及
    根据所述训练后的自编码结构的信息损失率从所述多个训练后的自编码结构中选取目标自编码结构。
  14. 根据权利要求11所述的计算机设备,其特征在于,所述处理器执行所述计算机可读指令时还执行以下步骤:
    将所述目标文本输入预设学习算法,输出目标输入向量;及
    将所述目标输入向量输入所述目标自编码结构,从所述目标自编码结构的中间层提取与所述目标输入向量对应的隐含特征向量。
  15. 根据权利要求13所述的计算机设备,其特征在于,所述处理器执行所述计算机可读指令时还执行以下步骤:
    将所述样本特征向量输入预存的自编码结构,得到对应的样本输出向量;
    获取所述样本特征向量和所述样本输出向量之间的信息损失量;及
    根据所述信息损失量调整所述自编码结构的参数,直至根据调整参数后的自编码结构得到的信息损失量符合预设条件,将所述调整参数后的自编码结构作为所述训练后的自编码结构。
  16. 一个或多个存储有计算机可读指令的非易失性计算机可读存储介质,所述计算机可读指令被一个或多个处理器执行时,使得所述一个或多个处理器执行以下步骤:
    获取输入的语音文本,并将所述语音文本转为目标文本;
    将所述目标文本输入预设的目标自编码结构,从所述目标自编码结构的中间层提取与所述目标文本对应的隐含特征向量;其中,所述目标自编码结构用于对所述目标文本进行编码并解码,以得到所述隐含特征向量;所述隐含特征向量用于表示所述目标文本的语义特征;
    获取与所述目标文本对应的预设的目标标准文本,将所述目标标准文本转化为目标标准向量;及
    获取所述隐含特征向量与所述目标标准向量之间的相似度,根据所述隐含特征向量和所述目标文本之间的对应关系,以及所述目标标准向量和所述目标标准文本之间的对应关系,将所述相似度作为所述目标文本和所述目标标准文本之间的匹配度。
  17. 根据权利要求16所述的存储介质,其特征在于,所述计算机可读指令被所述处理器执行时还执行以下步骤:在将所述目标文本输入预设的目标自编码结构,从所述目标自编码结构的中间层提取与所述目标文本对应的隐含特征向量之前,对多个预存的自编码结构分别进行训练,并从训练后的自编码结构中获取所述目标自编码结构。
  18. 根据权利要求17所述的存储介质,其特征在于,所述计算机可读指令被所述处理器执行时还执行以下步骤:
    将预存的多个样本特征向量输入预存的自编码结构,得到对应的样本输出向量;
    根据所述样本特征向量,对预存的多个自编码结构进行训练,得到多个训练后的自编码结构;及
    根据所述训练后的自编码结构的信息损失率从所述多个训练后的自编码结构中选取目标自编码结构。
  19. 根据权利要求16所述的存储介质,其特征在于,所述计算机可读指令被所述处理器执行时还执行以下步骤:
    将所述目标文本输入预设学习算法,输出目标输入向量;及
    将所述目标输入向量输入所述目标自编码结构,从所述目标自编码结构的中间层提取与所述目标输入向量对应的隐含特征向量。
  20. 根据权利要求18所述的存储介质,其特征在于,所述计算机可读指令被所述处理器执行时还执行以下步骤:
    将所述样本特征向量输入预存的自编码结构,得到对应的样本输出向量;
    获取所述样本特征向量和所述样本输出向量之间的信息损失量;及
    根据所述信息损失量调整所述自编码结构的参数,直至根据调整参数后的自编码结构得到的信息损失量符合预设条件,将所述调整参数后的自编码结构作为所述训练后的自编码结构。
PCT/CN2019/103474 2019-07-04 2019-08-30 文本匹配度检测方法、装置、计算机设备和可读存储介质 Ceased WO2021000412A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201910599137.6A CN110442677A (zh) 2019-07-04 2019-07-04 文本匹配度检测方法、装置、计算机设备和可读存储介质
CN201910599137.6 2019-07-04

Publications (1)

Publication Number Publication Date
WO2021000412A1 true WO2021000412A1 (zh) 2021-01-07

Family

ID=68429073

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2019/103474 Ceased WO2021000412A1 (zh) 2019-07-04 2019-08-30 文本匹配度检测方法、装置、计算机设备和可读存储介质

Country Status (2)

Country Link
CN (1) CN110442677A (zh)
WO (1) WO2021000412A1 (zh)

Cited By (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113254658A (zh) * 2021-07-07 2021-08-13 明品云(北京)数据科技有限公司 文本信息处理方法、系统、介质和设备
CN113486672A (zh) * 2021-07-27 2021-10-08 腾讯音乐娱乐科技(深圳)有限公司 多音字消歧方法及电子设备和计算机可读存储介质
CN115952264A (zh) * 2022-12-16 2023-04-11 浙江极氪智能科技有限公司 文本处理方法、装置和设备
CN116778011A (zh) * 2023-05-22 2023-09-19 阿里巴巴(中国)有限公司 图像生成方法
CN118570170A (zh) * 2024-06-03 2024-08-30 盐城工学院 一种基于多示例学习的类器官atp分析方法及系统

Families Citing this family (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111191028A (zh) * 2019-12-16 2020-05-22 浙江大搜车软件技术有限公司 样本标注方法、装置、计算机设备和存储介质
CN111368079B (zh) * 2020-02-28 2024-06-25 腾讯科技(深圳)有限公司 一种文本分类的方法、模型训练的方法、装置及存储介质

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20060190241A1 (en) * 2005-02-22 2006-08-24 Xerox Corporation Apparatus and methods for aligning words in bilingual sentences
CN108509407A (zh) * 2017-02-27 2018-09-07 广东神马搜索科技有限公司 文本语义相似度计算方法、装置及用户终端
US20190129947A1 (en) * 2017-10-26 2019-05-02 Electronics And Telecommunications Research Institute Neural machine translation method and apparatus
CN109766428A (zh) * 2019-02-02 2019-05-17 中国银行股份有限公司 数据查询方法和设备、数据处理方法
CN109840328A (zh) * 2019-02-28 2019-06-04 上海理工大学 深度学习商品评论文本情感倾向分析方法

Family Cites Families (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106126581B (zh) * 2016-06-20 2019-07-05 复旦大学 基于深度学习的手绘草图图像检索方法
CN109829299B (zh) * 2018-11-29 2022-05-10 电子科技大学 一种基于深度自编码器的未知攻击识别方法
CN109871531A (zh) * 2019-01-04 2019-06-11 平安科技(深圳)有限公司 隐含特征提取方法、装置、计算机设备及存储介质
CN109887562B (zh) * 2019-02-20 2021-10-29 广州天鹏计算机科技有限公司 电子病历的相似度确定方法、装置、设备和存储介质

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20060190241A1 (en) * 2005-02-22 2006-08-24 Xerox Corporation Apparatus and methods for aligning words in bilingual sentences
CN108509407A (zh) * 2017-02-27 2018-09-07 广东神马搜索科技有限公司 文本语义相似度计算方法、装置及用户终端
US20190129947A1 (en) * 2017-10-26 2019-05-02 Electronics And Telecommunications Research Institute Neural machine translation method and apparatus
CN109766428A (zh) * 2019-02-02 2019-05-17 中国银行股份有限公司 数据查询方法和设备、数据处理方法
CN109840328A (zh) * 2019-02-28 2019-06-04 上海理工大学 深度学习商品评论文本情感倾向分析方法

Cited By (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113254658A (zh) * 2021-07-07 2021-08-13 明品云(北京)数据科技有限公司 文本信息处理方法、系统、介质和设备
CN113486672A (zh) * 2021-07-27 2021-10-08 腾讯音乐娱乐科技(深圳)有限公司 多音字消歧方法及电子设备和计算机可读存储介质
CN115952264A (zh) * 2022-12-16 2023-04-11 浙江极氪智能科技有限公司 文本处理方法、装置和设备
CN116778011A (zh) * 2023-05-22 2023-09-19 阿里巴巴(中国)有限公司 图像生成方法
CN116778011B (zh) * 2023-05-22 2024-05-24 阿里巴巴(中国)有限公司 图像生成方法
CN118570170A (zh) * 2024-06-03 2024-08-30 盐城工学院 一种基于多示例学习的类器官atp分析方法及系统

Also Published As

Publication number Publication date
CN110442677A (zh) 2019-11-12

Similar Documents

Publication Publication Date Title
WO2020258506A1 (zh) 文本信息匹配度检测方法、装置、计算机设备和存储介质
WO2021000412A1 (zh) 文本匹配度检测方法、装置、计算机设备和可读存储介质
CN112766319B (zh) 对话意图识别模型训练方法、装置、计算机设备及介质
CN112668313B (zh) 句子智能纠错方法、装置、计算机设备及存储介质
CN110765763B (zh) 语音识别文本的纠错方法、装置、计算机设备和存储介质
WO2022142613A1 (zh) 训练语料扩充方法及装置、意图识别模型训练方法及装置
WO2021042503A1 (zh) 信息分类抽取方法、装置、计算机设备和存储介质
CN109271646A (zh) 文本翻译方法、装置、可读存储介质和计算机设备
WO2020177230A1 (zh) 基于机器学习的医疗数据分类方法、装置、计算机设备及存储介质
US20230123941A1 (en) Multiscale Quantization for Fast Similarity Search
WO2019080864A1 (zh) 一种文本语义编码方法及装置
WO2022257454A1 (zh) 一种合成语音的方法、装置、终端及存储介质
CN112380837B (zh) 基于翻译模型的相似句子匹配方法、装置、设备及介质
CN114973229A (zh) 文本识别模型训练、文本识别方法、装置、设备及介质
KR20230046946A (ko) 타겟 화자를 식별하는 전자 장치 및 이의 동작 방법
CN115410212B (zh) 多模态模型的训练方法、装置、计算机设备及存储介质
CN115495553A (zh) 查询文本排序方法、装置、计算机设备及存储介质
WO2025096348A1 (en) Intelligent knowledge-based question answering system
CN113836192B (zh) 平行语料的挖掘方法、装置、计算机设备及存储介质
CN109885811B (zh) 文章风格转换方法、装置、计算机设备及存储介质
CN115203372A (zh) 文本意图分类方法、装置、计算机设备及存储介质
CN116364061B (zh) 多场景语音识别方法、装置、计算机设备及存储介质
WO2023173547A1 (zh) 文本图像匹配方法、装置、设备及存储介质
CN115952767B (zh) 事件抽取方法、装置、设备及介质
CN118643819A (zh) 用于富文本内容查重的方法、处理器以及存储介质

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 19936519

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 19936519

Country of ref document: EP

Kind code of ref document: A1