A question-answer text matching method based on multi-layer semantic feature extraction structure

Abstract

The invention relates to the technical field of text recognition and matching, and discloses a question-answer text matching method based on a multi-layer semantic feature extraction structure, the question-answer text matching method based on the multi-layer semantic feature extraction structure comprising step one, control instruction confirmation; step two, voice information collection; step three, noise filtering; step four, instruction text voice matching; step five, instruction text voice recording. The question-answer text matching method based on the multi-layer semantic feature extraction structure pre-establishes a voice signal model, and recognizes and matches the collected and recognized instruction text through the voice signal model and the voice text instruction information pre-entered in the database, and at the same time sorts the command text from high to low according to the matching degree and the frequency of use, so that users can make choices according to actual needs, which can not only ensure the correctness of the instruction issuance, but also improve its intelligent interactive use effect.

Description

Technical Field

[0001] The present invention relates to the technical field of text recognition and matching, and in particular to a question-answering text matching method based on a multi-layer semantic feature extraction structure. Background Art

[0002] With the popularization of intelligent devices, voice intelligent terminals are gradually entering people's lives and work. By collecting voice information and translating voice text with voice intelligent terminals, command information can be obtained. Generally, intelligent voice terminals have autonomous learning functions, which automatically classify large amounts of text according to the semantics of the text, so as to better help people grasp the text information. The automatic text classification system is an important research direction of information processing. It refers to the process of automatically identifying the text category according to the content of the text under a given classification system. At present, methods in the fields of machine learning and data mining are used to process it. By classifying text information based on content, a text classification system that is easy for users to use is automatically generated, which can greatly reduce the human resources consumed in organizing documents and help users quickly find the information they need.

[0003] However, when existing intelligent information processing terminals collect voice information, since Chinese has many homophones, these homophones correspond to different words and meanings in different contexts, which to a certain extent affects the efficiency of intelligent information processing terminals when processing such text information and queries and executing certain instructions, resulting in errors when sending group SMS and text messages, affecting the efficiency and accuracy of communication. In view of this, we propose a question-answering text matching method based on a multi-layer semantic feature extraction structure. Summary of the invention

[0004] The purpose of the present invention is to provide a question-answer text matching method based on a multi-layer semantic feature extraction structure to solve the problems raised in the above background technology.

[0005] To achieve the above object, the present invention provides the following technical solution: a question-answer text matching method based on a multi-layer semantic feature extraction structure, which performs the following steps:

[0006] Step 1: Control command confirmation: When the intelligent information processing terminal is turned on, the control voice confirmation program is started to confirm the identity information by calculating the value of the matching degree between the input control command voice and the voice information in the database, and matching the corresponding voice command database at the same time;

[0007] Step 2: Collect voice information. The command information of the current environment is collected through the voice intelligent interaction module of the intelligent information processing terminal. The voice intelligent interaction module includes a voice collection device and an audio device, thereby realizing external voice text collection and internal text sound conversion and output, so as to realize voice intelligent interaction with the staff and facilitate intelligent voice control of the equipment;

[0008] Step 3: Filter the noise, using the filter device of the intelligent information processing terminal to filter out the noise other than the command text and voice, including the ambient sound and other text and voice information;

[0009] Step 4: Instruction text voice matching: the instruction text collected by the intelligent information processing terminal is matched and identified with the instruction text in the database, and the instruction text is sorted according to the selection rate of the instruction text in the database and the matching degree between the instruction text collected by the intelligent information processing terminal and the instruction text in the database, and finally displayed on the interactive screen of the intelligent information processing terminal;

[0010] Step 5: Instruction text voice recording. If no matching text information is retrieved based on step 3, the interactive screen of the intelligent information processing terminal can input the corresponding new text and choose to save it in the database.

[0011] Preferably, when performing voice recognition in step 1, the Bayesian formula is used to calculate the degree of matching between the control instruction and the voice signal and the identity information in the database of the intelligent information processing terminal, and finally the entered identity information is confirmed.

[0012] ,

[0013] Among them, H represents a specific text or command, and D represents a voice signal. The specific text or command is pre-recorded and uploaded to the database of the intelligent information processing terminal. In the initial use stage of the intelligent information processing terminal, a voice signal model is established for the specific voice signal through voice recognition calculation. When the intelligent information processing terminal is started in the subsequent use stage, the voice signal model is confirmed to facilitate subsequent voice information filtering and intelligent matching. At the same time, the number of the voice signal models can be set to multiple, and the higher the P (H|D) value, the higher the matching degree of the identity information, so as to enhance the flexibility of the intelligent information processing terminal in processing voice text information.

[0014] Preferably, the instruction information in the step 2 adopts an intelligent AI learning model, and the intelligent information processing terminal is connected to the Internet of Things. The corresponding contextual meanings are included when homophones and homophones are included. The specific form of the homophones and homophones and the corresponding contextual meaning information is A(X), and X is a constant. The sentence information corresponding to the corresponding contextual meanings of the homophones and homophones corresponds to Q+A(X)+W, wherein Q is the word text information appearing before the homophones and homophones, and W is the word text information appearing after the homophones and homophones, which is used as a reference when judging the matching degree of the instruction text to improve the recognition accuracy of the matching degree of the instruction text.

[0015] Preferably, the instruction information in step 2 includes daily instructions for daily use and control commands of the intelligent information processing terminal, and the daily instructions are entered into the database when the intelligent information processing terminal system is established, so that this part of the daily instructions can be matched and identified with the user's voice command information in a timely manner, making it convenient for the intelligent information processing terminal to respond in a timely manner.

[0016] Preferably, when filtering noise in step three, the voice signal model confirmed when the intelligent information processing terminal is turned on is used as a comparison basis, and the voice text information belonging to the current voice signal model is amplified during filtering, and the voice text information is captured through the text extraction model.

[0017] Preferably, when confirming the control command in step one, at most two information identities can be confirmed and selected at the same time, and a primary command identification identity and a secondary command identification identity need to be selected from the two information identities. When there is a conflict in voice commands, the primary command identification identity has priority, so that the staff can issue voice commands to control the intelligent information processing terminal according to actual needs, thereby facilitating intelligent equipment to assist people in their work.

[0018] Preferably, when performing instruction text matching, recognition and sorting in step 4, matching calculation is performed using a word accuracy calculation formula.

[0019] ,

[0020] Among them, the value of WCR represents the word accuracy rate, S represents the total number of words in the instruction text, D represents the number of incorrect words in the recognition result, I represents the number of new words in the recognition result, and the value of WCR∈(0,1). The closer the value of WCR is to 1, the more accurate the instruction text recognition is. This is to quantitatively evaluate the recognition accuracy and precision of speech and text, thereby improving the performance of the text-to-speech matching program. The WCR value of the instruction text matching recognition is used as label information to match the current instruction, and finally sorted from high to low according to the degree of matching.

[0021] Preferably, the judgment criteria for judging the contextual meaning information corresponding to the homophones and homophones include the recognition judgment of Q and W, and the frequency of retrieval and confirmation of the contextual meaning information in which the homophones and homophones are judged is arranged from high to low, and the retrieval and confirmation frequency is used as another judgment criterion for the priority in the subsequent judgment of homophones and homophones, so that homophones and homophones can be sorted according to the frequency of use during speech text recognition, which is convenient for users to instantly confirm correct and accurate text recognition information.

[0022] Preferably, in step four, the recognized voice text information is separately identified, matched and sorted according to different voice models when sorting, and the matching information is separately arranged and displayed on the interactive screen of the intelligent information processing terminal. When the instruction information is finally confirmed, the user of the corresponding voice model is required to confirm the text information, so that during the intelligent device control process, two login identities can use the voice text recognition system to perform instruction control of the intelligent information processing terminal at the same time, thereby improving the operating efficiency of the entire information processing system.

[0023] Preferably, in the step 3, a first-order filtering formula is used to perform filtering calculation when performing filtering processing.

[0024] ,

[0025] Among them, y(n) represents the current output signal value, x(n) represents the current input signal value, y(n-1) represents the output signal value at the previous moment, and a is a constant between 0 and 1, which determines the smoothness of the signal. By adjusting the value of a, different filtering effects can be achieved.

[0026] Compared with the prior art, the present invention provides a question-answer text matching method based on a multi-layer semantic feature extraction structure, which has the following beneficial effects:

[0027] 1. The question-answer text matching method based on the multi-layer semantic feature extraction structure establishes a voice signal model in advance, and recognizes and matches the collected and recognized command text through the voice signal model and the voice text command information pre-entered in the database, and sorts the command text from high to low according to the degree of matching and the frequency of use, so that users can choose according to their actual needs, thereby improving the effect of intelligent interactive use and ensuring the accuracy of text information when it is sent.

[0028] 2. The question-answer text matching method based on a multi-layer semantic feature extraction structure adopts an intelligent AI learning model to learn and match homophones and homophones, and refers to the sentence information corresponding to the corresponding contextual meanings of the homophones and homophones, as well as the preceding word text information and the succeeding word text information when recognizing and matching voice text instructions, and uses this as a reference when judging the instruction text matching degree, so as to improve the recognition accuracy of the instruction text matching degree, thereby ensuring the accuracy of the text information when it is sent. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] Figure 1 This is a schematic diagram of the text instruction matching process of the present invention;

[0030] Figure 2 The figure is a flow chart of the text instruction recognition and matching process of the present invention. DETAILED DESCRIPTION

[0031] like Figure 1-Figure 2 As shown, the present invention provides a technical solution: a question-answer text matching method based on a multi-layer semantic feature extraction structure, which performs the following steps:

[0032] Step 1: Control command confirmation: When the intelligent information processing terminal is turned on, the control voice confirmation program is started to confirm the identity information by calculating the value of the matching degree between the input control command voice and the voice information in the database, and matching the corresponding voice command database at the same time;

[0033] Step 2: Collect voice information. The command information of the current environment is collected through the voice intelligent interaction module of the intelligent information processing terminal. The voice intelligent interaction module includes a voice collection device and an audio device, thereby realizing external voice text collection and internal text sound conversion and output, so as to realize voice intelligent interaction with the staff and facilitate intelligent voice control of the equipment;

[0034] Step 3: Filter the noise, using the filter device of the intelligent information processing terminal to filter out the noise other than the command text and voice, including the ambient sound and other text and voice information;

[0035] Step 4: Instruction text voice matching: the instruction text collected by the intelligent information processing terminal is matched and identified with the instruction text in the database, and the instruction text is sorted according to the selection rate of the instruction text in the database and the matching degree between the instruction text collected by the intelligent information processing terminal and the instruction text in the database, and finally displayed on the interactive screen of the intelligent information processing terminal;

[0036] Step 5: Instruction text voice recording. If no matching text information is retrieved based on step 3, the interactive screen of the intelligent information processing terminal can input the corresponding new text and choose to save it in the database.

[0037] In one embodiment of the present invention, when performing voice recognition in step 1, the Bayesian formula is used to calculate the degree of matching between the control instruction and the voice signal and the identity information in the database of the intelligent information processing terminal, and finally confirm the entered identity information.

[0038] ,

[0039] Among them, H represents a specific text or command, and D represents a voice signal. The specific text or command is pre-recorded and uploaded and stored in the database of the intelligent information processing terminal. In the initial use stage of the intelligent information processing terminal, a voice signal model is established for the specific voice signal through voice recognition calculation. When the intelligent information processing terminal is started in the subsequent use stage, the voice signal model is confirmed to facilitate subsequent voice information filtering and intelligent matching. At the same time, the number of voice signal models can be set to multiple to enhance the flexibility of the intelligent information processing terminal in processing voice text information, and the higher the P (H | D) value, the higher the matching degree of identity information. At the same time, the instruction information in step 2 includes daily instructions for daily use and control commands of the intelligent information processing terminal, and the daily instructions are entered into the database when the intelligent information processing terminal system is established, so that this part of the daily instructions can be timely matched and identified with the user's voice instruction information, so that the intelligent information processing terminal can respond in time.

[0040] In addition, the instruction information in step 2 adopts an intelligent AI learning model, and the intelligent information processing terminal is connected to the Internet of Things. The corresponding contextual meanings are included when homophones and homophones are included. The specific form of the homophones and homophones and the corresponding contextual meaning information is A(X), and X is a constant. The sentence information corresponding to the corresponding contextual meanings of the homophones and homophones corresponds to Q+A(X)+W, where Q is the word text information appearing before the homophone and homophone, and W is the word text information appearing after the homophone and homophone, which is used as a reference when judging the matching degree of the instruction text to improve the recognition accuracy of the instruction text matching degree.

[0041] Specifically, the judgment criteria for judging the contextual meaning information corresponding to homophones and homophones include the recognition judgment of Q and W, and the frequency of retrieval and confirmation of the contextual meaning information in which the homophones and homophones are judged is arranged from high to low, and the retrieval and confirmation frequency is used as another judgment criterion for the priority in the subsequent judgment of homophones and homophones, so that homophones and homophones can be sorted according to the frequency of use during speech text recognition, which is convenient for use and immediate confirmation of correct and accurate text recognition information.

[0042] In an embodiment of the present invention, when filtering noise in step 3, the voice signal model confirmed when the intelligent information processing terminal is turned on is used as a comparison basis, and the voice text information belonging to the current voice signal model is amplified during filtering, and the voice text information is captured through the text extraction model, so that the recognition text information extracted from the intelligent recognition terminal can be arranged in order, thereby facilitating the user to select according to actual needs. At the same time, the first-order filtering formula is used for filtering calculation during filtering processing in step 3.

[0043] ,

[0044] Among them, y(n) represents the current output signal value, x(n) represents the current input signal value, y(n-1) represents the output signal value at the previous moment, and a is a constant between 0 and 1, which determines the smoothness of the signal. By adjusting the value of a, different filtering effects can be achieved.

[0045] Furthermore, when confirming the control command in step one, at most two information identities can be confirmed and selected at the same time, and the main command identification identity and the secondary command identification identity need to be selected from the two information identities. When there is a conflict in voice commands, the main command identification identity has priority, so that the staff can issue voice commands to control the intelligent information processing terminal according to actual needs, thereby facilitating intelligent equipment to assist people in their work.

[0046] In the present invention, when the instruction text matching recognition and sorting are performed in step 4, the matching calculation is performed by the word accuracy calculation formula:

[0047]

[0048] Among them, the value of WCR represents the word accuracy rate, S represents the total number of words in the instruction text, D represents the number of incorrect words in the recognition result, I represents the number of new words in the recognition result, and the value of WCR∈(0,1). The closer the value of WCR is to 1, the more accurate the instruction text recognition is. This is to quantitatively evaluate the recognition accuracy and precision of speech and text, thereby improving the performance of the text-to-speech matching program. The WCR value of the instruction text matching recognition is used as label information to match the current instruction, and finally sorted from high to low according to the degree of matching.

[0049] In addition, in step four, the recognized voice text information is separately identified, matched and sorted according to different voice models during sorting, and the matching information is separately arranged and displayed on the interactive screen of the intelligent information processing terminal. When the command information is finally confirmed, the user of the corresponding voice model is required to confirm the text information. In this way, during the intelligent device control process, two login identities can use the voice text recognition system to perform command control of the intelligent information processing terminal at the same time, thereby improving the operating efficiency of the entire information processing system.

[0050] The above generally describes the present invention in detail, but it is obvious to a person skilled in the art that some modifications or improvements can be made to the present invention. Therefore, modifications or improvements that do not depart from the spirit of the present invention are within the scope of protection of the present invention.

Claims

1. A question-answer text matching method based on a multi-layer semantic feature extraction structure, characterized in that: It performs the following steps: Step 1: Control command confirmation: When the intelligent information processing terminal is turned on, the control voice confirmation program is started to confirm the identity information by calculating the value of the matching degree between the input control command voice and the voice information in the database, and matching the corresponding voice command database at the same time; Step 2: Collect voice information, and collect the command information of the current environment through the voice intelligent interaction module of the intelligent information processing terminal, wherein the voice intelligent interaction module includes a voice collection device and an audio device; Step 3: Filter the noise, using the filter device of the intelligent information processing terminal to filter out the noise other than the command text and voice, including the ambient sound and other text and voice information; Step 4: Instruction text voice matching: the instruction text collected by the intelligent information processing terminal is matched and identified with the instruction text in the database, and the instruction text is sorted according to the selection rate of the instruction text in the database and the matching degree between the instruction text collected by the intelligent information processing terminal and the instruction text in the database, and finally displayed on the interactive screen of the intelligent information processing terminal; Step 5: Instruction text voice recording. If no matching text information is retrieved based on step 3, the intelligent information processing terminal interactive screen can input the corresponding new text and choose to save it in the database; The instruction information in the step 2 adopts an intelligent AI learning model, and the intelligent information processing terminal is connected to the Internet of Things. When homophones and homophones are collected, the corresponding contextual meanings are included. The specific form of the homophones and homophones and the corresponding contextual meaning information is A(X), and X is a constant. The sentence information corresponding to the corresponding contextual meaning of the homophones and homophones corresponds to Q+A(X)+W, wherein Q is the word text information appearing before the homophones and homophones, and W is the word text information appearing after the homophones and homophones, which is used as a reference when judging the matching degree of the instruction text. The judgment criteria for judging the corresponding contextual meaning information of the homophones and homophones include the recognition judgment of Q and W, and the frequency of retrieving and confirming the contextual meaning information in which the homophones and homophones are judged is arranged from high to low, and the retrieval confirmation frequency is used as another judgment criterion for the priority in the subsequent judgment of homophones and homophones.

2. The question-answer text matching method based on a multi-layer semantic feature extraction structure according to claim 1, characterized in that: When performing voice recognition in step 1, the Bayesian formula is used to calculate the degree of matching between the control command and the voice signal and the identity information in the database of the intelligent information processing terminal, and finally the entered identity information is confirmed. ， Among them, H represents a specific text or command, and D represents a voice signal. The specific text or command is pre-recorded and uploaded to the database of the intelligent information processing terminal. In the initial use stage of the intelligent information processing terminal, a voice signal model is established for the specific voice signal through voice recognition calculation. When the intelligent information processing terminal is started in the subsequent use stage, the voice signal model is confirmed. At the same time, the number of the voice signal models is set to multiple, and the higher the P (H|D) value, the higher the matching degree of the identity information.

3. The question-answer text matching method based on a multi-layer semantic feature extraction structure according to claim 2, characterized in that: The instruction information in step 2 includes daily instructions for daily use and control commands of the intelligent information processing terminal, and the daily instructions are entered into the database when the intelligent information processing terminal system is established.

4. The question-answer text matching method based on a multi-layer semantic feature extraction structure according to claim 2, characterized in that: When filtering noise in step three, the voice signal model confirmed when the intelligent information processing terminal is turned on is used as a comparison basis, and the voice text information belonging to the current voice signal model is amplified during filtering, and the voice text information is captured through the text extraction model.

5. The question-answer text matching method based on a multi-layer semantic feature extraction structure according to claim 2, characterized in that: When confirming the control command in step 1, at most two information identities can be confirmed and selected at the same time, and a primary command identification identity and a secondary command identification identity need to be selected from the two information identities. When there is a conflict in voice commands, the primary command identification identity has priority.

6. The question-answer text matching method based on a multi-layer semantic feature extraction structure according to claim 1, characterized in that: When performing instruction text matching, recognition and sorting in step 4, matching calculation is performed using the word accuracy calculation formula: ， Among them, the value of WCR represents the word accuracy rate, S represents the total number of words in the instruction text, D represents the number of incorrect words in the recognition result, I represents the number of new words in the recognition result, the value of WCR∈(0,1), and the closer the value of WCR is to 1, the more accurate the instruction text recognition is. The WCR value of the instruction text matching recognition is used as label information to match the current instruction, and finally sorted from high to low according to the degree of matching.

7. The question-answer text matching method based on a multi-layer semantic feature extraction structure according to claim 5, characterized in that: In step 4, the recognized voice text information is separately identified, matched and sorted according to different voice models during sorting, and the matching information is separately arranged and displayed on the interactive screen of the intelligent information processing terminal. When the instruction information is finally confirmed, the user of the corresponding voice model is required to confirm the text information.

8. The question-answer text matching method based on a multi-layer semantic feature extraction structure according to claim 4, characterized in that: When performing filtering processing in step 3, a first-order filtering formula is used for filtering calculation. ， Among them, y(n) represents the current output signal value, x(n) represents the current input signal value, y(n-1) represents the output signal value at the previous moment, and a is a constant between 0 and 1, which determines the smoothness of the signal. By adjusting the value of a, different filtering effects can be achieved.

Images