AI question and answer interaction method based on television voice remote controller

Abstract

The invention relates to the technical field of data retrieval, and discloses an AI question and answer interaction method based on a television voice remote controller. Comprising the following steps: extracting request features from request data, identifying compliance states of the request features, dividing the request features into retrieval features and execution features, screening out to-be-identified data from an AI database, calculating a matching degree of the to-be-identified data, marking target data, formulating a data interaction instruction according to the execution features, and executing the data interaction instruction. Identifying the priority of the data interaction instruction; compared with the prior art, illegal and unreasonable AI question and answer interaction requests can be discriminated at the source, the threshold of AI question and answer interaction is improved, the workload of AI question and answer interaction is reduced, and meanwhile, by means of the operation of dividing the request features into the retrieval features and the execution features, the accuracy of the AI question and answer interaction is improved. It is ensured that the AI server can be matched with the target data according to the retrieval features, the target data are executed in order, and the fluency and orderliness of AI question and answer interaction are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of data retrieval, and more specifically, to an AI question-answering interaction method based on a television voice remote controller. Background Art

[0002] With the rapid development of smart home and artificial intelligence technology, users have put forward higher requirements for the control methods of home entertainment devices. Against this background, AI voice interaction technology has gradually emerged and become an important means to enhance users' TV experience. Through voice recognition and natural language processing technology, users can easily implement AI question-and-answer interaction operations such as channel switching and program search through language commands.

[0003] The patent application with reference publication number CN116244417A discloses a question-answer interaction data processing method and server for AI chatbots, including performing text vector mining on the consultation text to be analyzed of the second dialogue type through a debugged second Transformer network to obtain a text description vector of the consultation text to be analyzed; wherein the consultation text to be analyzed is used to characterize the question-answer interaction record between the intelligent terminal and the chatbot; the question-answer strategy of the chatbot is upgraded and optimized based on the text description vector of the consultation text to be analyzed; it obtains a transfer reference vector of the second Transformer network by calling the consultation text example of the first dialogue type and the first Transformer network for collaborative debugging and analysis, and debugs the second Transformer network based on the transfer reference vector and the consultation text example of the second dialogue type, thereby combining the extraction of the transfer reference vector to ensure the transfer debugging quality when the consultation text example is insufficient, so as to improve the debugging timeliness of the Transformer network and reduce unnecessary time and computing expenses;

[0004] The existing AI question-and-answer interaction method uses a method of receiving voice data in real time and analyzing it word by word to extract the key features of the voice data, and gives retrieval data that matches the key features to achieve the AI ​​question-and-answer interaction effect. Due to the lack of legality and rationality screening and judgment operations for the input voice data, when a large number of features and illegal or unreasonable voice data are received, these abnormal voice data will cause confusion or even errors in the data retrieval of the AI ​​question-and-answer interaction, which not only lowers the threshold of AI question-and-answer interaction, but also reduces the fluency and orderliness of AI question-and-answer interaction.

[0005] In view of this, the present invention proposes an AI question-answering interaction method based on a TV voice remote control to solve the above problems. Summary of the invention

[0006] In order to overcome the above-mentioned defects of the prior art and to achieve the above-mentioned purpose, the present invention provides the following technical solution: an AI question-answering interaction method based on a TV voice remote control, applied to an AI server, comprising:

[0007] S1: receiving request data from the remote control end, and extracting request features from the request data based on a feature extraction criterion, wherein the feature extraction criterion is: taking the first character as the start position and the last character as the end position, and performing feature extraction in sequence;

[0008] S2: Based on the compliance criteria, the request features are identified for compliance, the compliance status of the request features is identified, and it is determined whether to enter the question-and-answer interaction mode. The compliance status includes valid compliance and invalid compliance. If the question-and-answer interaction mode is entered, S3 is executed; if the question-and-answer interaction mode is not entered, S1-S2 are repeated.

[0009] S3: identifying the feature attributes of the request features one by one, the feature attributes include noun attributes and verb attributes, and dividing the request features into retrieval features and execution features according to the feature attributes;

[0010] S4: Filter out the data to be identified that matches the search feature from the AI ​​database, calculate the matching degree of the data to be identified, and mark the target data from the data to be identified;

[0011] S5: Formulate data interaction instructions according to the execution characteristics, identify the priority of the data interaction instructions, and control the target data on the TV side to execute the data interaction instructions.

[0012] Furthermore, the method for extracting the request feature includes:

[0013] Receive the voice information of the personnel in real time through the remote control terminal, and record the time of first receiving the voice information as the starting time;

[0014] When no more voice information is received within the preset receiving time, the time when the voice information is last received is recorded as the end time, and the voice information received during the period from the start time to the end time is recorded as the request data;

[0015] Mark all characters in the request data one by one, and arrange all characters in order according to the time of receipt to generate a character row;

[0016] Taking the first character in the character row as the starting point, the key characters in the character row are identified one by one through natural language processing technology, and the association status of the key characters in adjacent positions is identified through the voice model;

[0017] Taking the last character in the character row as the end point, i key characters with associated status are combined head to tail to obtain i request features.

[0018] Furthermore, the compliance criteria are: the number of abnormal characters in the compliant request features is less than or equal to 1;

[0019] Methods for identifying effective and ineffective compliance include:

[0020] Mark the characters of i requested features one by one, and compare the characters of i requested features with the compliant characters in the compliant database one by one for consistency;

[0021] Characters that are inconsistent with the compliant characters in the compliant database are recorded as abnormal characters, and the number of abnormal characters in i request features is counted one by one to obtain i abnormal values;

[0022] The compliance status of the request feature with an outlier value less than or equal to 1 is recorded as valid compliance;

[0023] The compliance status of the request feature with an outlier value greater than 1 is recorded as invalid compliance.

[0024] Furthermore, the method for determining whether to enter the question-and-answer interaction mode includes:

[0025] Counting the number of request features corresponding to effective compliance and the number of request features corresponding to invalid compliance one by one, and obtaining a first compliance value and a second compliance value respectively;

[0026] comparing the first compliance value and the second compliance value to generate a compliance ratio;

[0027] The expression of the compliance ratio is:

[0028]

[0029] In the formula, HG bz is the compliance ratio, HG d1 is the first compliance value, HG d2 is the second compliance value;

[0030] The compliance ratio HG bz Compared with the preset compliance threshold HG yz Compare;

[0031] When HG bz Greater than or equal to HG yz When , it is determined to enter the question-answer interaction mode;

[0032] When HG bz Smaller than HG yz , it is determined not to enter the question-and-answer interactive mode.

[0033] Furthermore, the identification method of noun attributes and verb attributes includes:

[0034] Arrange the i request features in chronological order, and repeatedly compare the characters of the i request features with the words in the noun database and the verb database in turn;

[0035] When the characters of the requested feature are repeated with words in the noun database, the feature attribute of the requested feature is identified as a noun attribute;

[0036] When characters of the request feature are repeated with words in the verb database, the feature attribute of the request feature is identified as a verb attribute.

[0037] Furthermore, the division method of the retrieval feature and the execution feature includes:

[0038] When the feature attribute of the request feature is a noun attribute, the request feature is divided into retrieval features to obtain p retrieval features;

[0039] When the feature attribute of the request feature is a verb attribute, the request feature is divided into execution features to obtain m execution features.

[0040] Furthermore, the matching degree calculation method includes:

[0041] Using natural language processing technology, the first keywords of w stored data in the AI ​​database are identified one by one to obtain w first keywords;

[0042] The first keyword repeated with any search feature is recorded as the target keyword, and the stored data corresponding to the target keyword is recorded as the data to be identified, and X data to be identified are obtained;

[0043] Through the attribute management system, X attribute boxes of the data to be identified are searched one by one, the remark words in the attribute boxes are marked, and the number of the remark words is counted to obtain X total values;

[0044] The remark words that are the same as the search features are recorded as valid remark, and the number of valid remark in the X data to be identified is counted to obtain X remark values;

[0045] Compare X remark values ​​with X total values ​​one by one to obtain X matching degrees;

[0046] The expression of matching degree is:

[0047]

[0048] In the formula, PD dx is the matching degree of the Xth data to be identified, BZ zx is the remark value of the xth data to be identified, ZL zx is the total value of the xth data to be identified.

[0049] Furthermore, the target data marking method includes:

[0050] Arrange the X pieces of data to be identified in order from largest to smallest matching degree;

[0051] When the matching degrees of the first, second and third to-be-identified data are all inconsistent, the first, second and third to-be-identified data are recorded as target data;

[0052] When the matching degrees of the first, second and third data to be identified are the same, the multiple data to be identified that are located in the first, second or third positions are arranged in descending order according to the remark values ​​from large to small to obtain the data arrangement rows in descending order;

[0053] The data to be identified at the first, second and third positions in the descending data arrangement row are recorded as target data.

[0054] Further, the data interaction instructions include form determination instructions, mode selection instructions, and time jump instructions;

[0055] The formulation method of the form determination instruction, the mode selection instruction and the time jump instruction includes:

[0056] Identify the second keywords of the execution features one by one through natural language processing technology to obtain m second keywords;

[0057] Record the form structure, form specification and form attribute as the first instruction word;

[0058] Recording the working mode, mode type and mode attribute as the second instruction word;

[0059] Record time jump, time switch and time attribute as the third instruction word;

[0060] When the first instruction word exists in the m second keywords, a form determination instruction is formulated;

[0061] When the second instruction word exists in the m second keywords, a mode selection instruction is formulated;

[0062] When the third instruction word exists among the m second keywords, a time jump instruction is formulated.

[0063] Furthermore, the priority is: the priority of the form determination instruction is higher than the priority of the mode selection instruction, and the priority of the mode selection instruction is higher than the priority of the time jump instruction.

[0064] The technical effects and advantages of the AI ​​question-answering interactive method based on a TV voice remote controller of the present invention are as follows:

[0065] The present invention receives request data from a remote control terminal, extracts request features from the request data based on feature extraction criteria, performs compliance identification on the request features based on compliance criteria, identifies the compliance status of the request features, determines whether to enter a question-and-answer interaction mode, identifies feature attributes of the request features one by one, and divides the request features into retrieval features and execution features according to the feature attributes, selects data to be identified that matches the retrieval features from an AI database, calculates the matching degree of the data to be identified, marks target data from the data to be identified, formulates data interaction instructions based on the execution features, identifies the priority of the data interaction instructions, and controls the target data on the TV terminal to execute the data interaction instructions; compared with the prior art, based on the feature extraction criteria, the present invention can It can extract request features from the received request data and identify whether the request features meet the AI ​​question-answering interaction requirements, so that it can identify illegal and unreasonable AI question-answering interaction requests at the source, which not only raises the threshold for AI question-answering interaction, but also reduces the workload of AI question-answering interaction. At the same time, it divides the request features into retrieval features and execution features to ensure that the AI ​​server can match the target data according to the retrieval features, and execute the target data in an orderly manner through the TV end based on the execution features, thereby ensuring that the target data corresponding to the request data can be executed accurately and orderly, improving the fluency and orderliness of AI question-answering interaction, and meeting people's high-quality requirements for AI question-answering interaction through the TV voice remote control. BRIEF DESCRIPTION OF THE DRAWINGS

[0066] Figure 1 A flowchart of an AI question-answering interactive method based on a TV voice remote controller provided in Embodiment 1 of the present invention;

[0067] Figure 2 A schematic diagram of an AI question-answering interactive system based on a TV voice remote control provided in Embodiment 2 of the present invention. DETAILED DESCRIPTION

[0068] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0069] Example 1: Please refer to Figure 1 As shown, the AI ​​question-answering interaction method based on a TV voice remote control described in this embodiment is applied to an AI server, including:

[0070] S1: receiving request data from the remote control end, and extracting request features from the request data based on feature extraction criteria;

[0071] The remote control end refers to a port that can provide demand information to the AI ​​server, serving as the input end for AI question-and-answer interaction. Specifically, the remote control end is a TV voice remote control, and the TV voice remote control should be equipped with a microphone and other related voice input modules. The request data refers to the voice demand data input by the remote control end to the AI ​​server, which can be used as the pre-data for the remote control end to perform AI question-and-answer interaction.

[0072] Since the request data is raw data received by the remote control terminal and has not been analyzed and processed, when the request data contains a lot of complicated information, the AI ​​server cannot accurately identify the true meaning of the request data at the first time. Therefore, it is necessary to analyze the request data and extract the request features that can express the true meaning of the request data;

[0073] When extracting request features, it is necessary to ensure that the request features can be extracted in an orderly and accurate manner without extraction errors or disorder. Therefore, it is necessary to restrict the extraction process of request features through feature extraction criteria.

[0074] The feature extraction criteria are: take the first character as the starting position and the last character as the ending position, and perform feature extraction in sequence;

[0075] The methods for extracting request features include:

[0076] Receive the voice information of the personnel in real time through the remote control terminal, and record the time of first receiving the voice information as the starting time;

[0077] When no voice information is received within the preset receiving time, the time when the voice information is last received is recorded as the end time, and the voice information received during the period from the start time to the end time is recorded as the request data; the preset receiving time refers to the maximum time span of the voice information of the remote control receiving person, which can limit the acquisition time span of the request data, so that the request data within the reasonable receiving time can be fully received, thereby improving the accuracy of receiving the request data;

[0078] Mark all characters in the request data one by one, and arrange all characters in order according to the time of receipt to generate a character row;

[0079] Taking the first character in the character row as the starting point, the key characters in the character row are identified one by one through natural language processing technology, and the association status of the key characters in adjacent positions is identified through the speech model; the key characters refer to the characters in the character row that can correspond to the pre-stored words with the requested real meaning expression, so that the characters with real meaning expression can be accurately identified from a large number of characters, and the association status is used to indicate the degree of association between the key characters in adjacent positions whether they represent the same real meaning, that is, multiple key characters representing the same real meaning can be effectively identified; the association status includes association and non-association;

[0080] Taking the last character in the character row as the end point, i key characters with associated status are combined head to tail to obtain i request features.

[0081] It should be noted that the number of request features in the request data is not unique, but is determined according to the number of different meanings expressed by the request data, and each request feature corresponds to a separate meaning. There will not be a phenomenon where two request features represent the same meaning, thereby ensuring the independence of each request feature.

[0082] S2: Based on the compliance criteria, the request features are identified for compliance, the compliance status of the request features is identified, and it is determined whether to enter the question-and-answer interaction mode;

[0083] Compliance identification refers to the identification operation of whether the request characteristics conform to the pre-set principles of AI question-answering interaction, thereby identifying the compliance of each request feature obtained;

[0084] After the compliance identification of the request features is performed, the identification result of each request feature, i.e., the compliance status, can be obtained; the compliance status includes effective compliance and invalid compliance; effective compliance means that the request feature has passed the compliance identification, and the request feature complies with the pre-set principle; invalid compliance means that the request feature has not passed the compliance identification, and the request feature does not comply with the pre-set principle;

[0085] When conducting compliance identification, the principle for determining compliance is the compliance criteria, so that the compliance criteria can serve as the basis for subsequent compliance identification and identification of compliance status;

[0086] The compliance criterion is: the number of abnormal characters in the compliant request feature is less than or equal to 1; that is, the request feature containing too many non-compliant characters can be effectively identified, thereby avoiding the request feature from retrieving a large amount of non-compliant data information;

[0087] Methods for identifying effective and ineffective compliance include:

[0088] Mark the characters of i request features one by one, and compare the characters of i request features with the compliant characters in the compliant database one by one for consistency; the compliant database refers to a database composed of a large number of compliant characters, so that the characters in the compliant database are in a compliant state, thereby providing a comparison basis for the subsequent recognition and judgment of abnormal characters;

[0089] Characters that are inconsistent with the compliant characters in the compliant database are recorded as abnormal characters, and the number of abnormal characters in i request features is counted one by one to obtain i abnormal values;

[0090] The compliance status of the request feature with an outlier value less than or equal to 1 is recorded as valid compliance;

[0091] The compliance status of the request feature with an outlier value greater than 1 is recorded as invalid compliance.

[0092] After obtaining the compliance status of the request feature, it can be determined whether the request data of the remote control end can perform the next question-and-answer interaction operation, that is, whether to enter the question-and-answer interaction mode, based on the number of request features corresponding to valid compliance and request features corresponding to invalid compliance;

[0093] When the number of request features corresponding to effective compliance and the number of request features corresponding to invalid compliance meet the minimum standard for entering the question-and-answer interaction mode after comparison, the question-and-answer interaction mode needs to be entered at this time, so that subsequent steps can be executed in the question-and-answer interaction mode;

[0094] Methods for determining whether to enter the question-and-answer interactive mode include:

[0095] Counting the number of request features corresponding to effective compliance and the number of request features corresponding to invalid compliance one by one, and obtaining a first compliance value and a second compliance value respectively;

[0096] comparing the first compliance value and the second compliance value to generate a compliance ratio;

[0097] The expression of the compliance ratio is:

[0098]

[0099] In the formula, HG bz is the compliance ratio, HG d1 is the first compliance value, HG d2 is the second compliance value;

[0100] The compliance ratio HG bz Compared with the preset compliance threshold HG yzComparison; The preset compliance threshold is a numerical representation of the minimum value of the compliance ratio corresponding to the question-answer interaction mode, so that the size of the calculated compliance ratio can be distinguished, and whether the question-answer interaction mode is entered can be accurately identified, and then the overall compliance of the request data of the remote control end can be identified, preventing non-compliant request data from entering the AI ​​server for subsequent question-answer interaction operations, thereby avoiding the occurrence of illegal and unreasonable data information; The preset compliance threshold is obtained by collecting a large number of historical minimum values ​​of the compliance ratio corresponding to entering the question-answer interaction mode and calculating the average value;

[0101] When HG bz Greater than or equal to HG yz When , it means that the compliance ratio is greater than or equal to the preset compliance threshold. At this time, the number of valid and compliant request features is large, and the request data will not have illegal anomalies during question-and-answer interaction, then it is determined to enter the question-and-answer interaction mode;

[0102] When HG bz Smaller than HG yz When , it means that the compliance ratio is less than the preset compliance threshold. At this time, the number of valid and compliant request features is small, and illegal anomalies may occur in the request data during question-and-answer interaction. In this case, it is determined not to enter the question-and-answer interaction mode.

[0103] It should be noted that when it is determined that the question-and-answer interaction mode is entered, the subsequent steps S3-S5 can be continued to be executed. At this time, the user can establish a two-way interactive channel with the AI ​​server through the remote control and achieve the effect of question-and-answer interaction. When it is determined that the question-and-answer interaction mode is not entered, the subsequent steps S3-S5 cannot be executed. At this time, the AI ​​server will not provide the user with subsequent question-and-answer interaction services, and it is necessary to return to the step S1 and execute S2 again to make a judgment until the question-and-answer interaction mode can be entered.

[0104] S3: identifying the feature attributes of the request features one by one, and dividing the request features into retrieval features and execution features according to the feature attributes;

[0105] Feature attributes refer to the representation of different attributes of the data corresponding to the request features, so that the request features with different feature attributes can meet different AI question-answering interaction requirements, and according to the different feature attributes, the corresponding operations can be screened and executed from the AI ​​database;

[0106] Feature attributes include noun attributes and verb attributes. Noun attributes refer to the attributes of the requested features that belong to nouns and serve as the basis for the initial screening and matching of subsequent data. Verb attributes refer to the attributes of the requested features that belong to verbs and serve as the basis for subsequent data execution actions.

[0107] The identification methods of noun attributes and verb attributes include:

[0108] Arrange the i request features in chronological order, and repeatedly compare the characters of the i request features with the words in the noun database and the verb database in turn;

[0109] When the characters of the requested feature are repeated with words in the noun database, it means that the requested feature is used for data representation of a noun, and the feature attribute of the requested feature is identified as a noun attribute;

[0110] When characters of the request feature are repeated with words in the verb database, it indicates that the request feature is used for data representation of a verb, and the feature attribute of the request feature is identified as a verb attribute.

[0111] After obtaining the characteristic attributes of the request features, the request features can be effectively divided according to the different characteristic attributes, so that the request features corresponding to nouns and the request features corresponding to verbs are used for data matching and execution respectively, and then divided into retrieval features and execution features;

[0112] The division methods of retrieval features and execution features include:

[0113] When the feature attribute of the request feature is a noun attribute, the request feature is used as the basis for subsequent data retrieval and matching, and the request feature is divided into retrieval features to obtain p retrieval features;

[0114] When the feature attribute of the request feature is a verb attribute, the request feature is used as a basis for subsequent data execution operations, and the request feature is divided into execution features to obtain m execution features.

[0115] It should be noted that the number of retrieval features and the number of execution features are both greater than or equal to 1, thereby ensuring that the data corresponding to the request data can be retrieved and executed, and further ensuring that the retrieved and matched data can be effectively and accurately executed.

[0116] S4: Filter out the data to be identified that matches the search feature from the AI ​​database, calculate the matching degree of the data to be identified, and mark the target data from the data to be identified;

[0117] The data to be identified refers to the corresponding data retrieved from the AI ​​database using the search feature as the screening criteria, so that the data to be identified can be effectively matched with the search feature and is used as the original data provided by the AI ​​server to the remote control end;

[0118] Since there is a lot of data stored in the AI ​​database, and the amount of data that can match and associate with the search features is also large, the amount of data to be identified is large. Since the matching degree of the data to be identified and the search features is different, it is necessary to calculate the matching degree of the data to be identified with different matching degrees;

[0119] The matching degree is calculated by:

[0120] The first keywords of w stored data in the AI ​​database are identified one by one through natural language processing technology to obtain w first keywords; the first keywords are words used to represent the true meaning of the data stored in the AI ​​database, so that the stored data with more and more complicated characters can be represented concisely and directly;

[0121] The first keyword repeated with any search feature is recorded as the target keyword, and the stored data corresponding to the target keyword is recorded as the data to be identified, and X data to be identified are obtained;

[0122] Through the attribute management system, X attribute boxes of the data to be identified are queried one by one, the remark words in the attribute boxes are marked, and the number of remark words is counted to obtain X total values; the attribute box is used to express the overall outline of the remark content of the data to be identified, so as to ensure that the remark content of the data to be identified can be effectively summarized together, and the remark word is the smallest unit constituting the attribute box, and is used to directly express the remark content;

[0123] The remark words that are the same as the search features are recorded as valid remark, and the number of valid remark in the X data to be identified is counted to obtain X remark values;

[0124] Compare X remark values ​​with X total values ​​one by one to obtain X matching degrees;

[0125] The expression of matching degree is:

[0126]

[0127] In the formula, PD dx is the matching degree of the xth data to be identified, BZ zx is the remark value of the Xth data to be identified, ZL zx is the total value of the Xth data to be identified.

[0128] After obtaining the matching degree of the data to be identified, the data to be identified can be re-identified and screened according to the matching degree, and the re-identified and screened data to be identified is recorded as the target data. The target data at this time can be used as data that matches the request data of the remote control end and provide data objects for subsequent AI question-and-answer interactions;

[0129] The target data labeling methods include:

[0130] Arrange the X pieces of data to be identified in order from largest to smallest matching degree;

[0131] When the matching degrees of the first, second and third to-be-identified data are all inconsistent, the first, second and third to-be-identified data are recorded as target data;

[0132] When the matching degrees of the data to be identified at the first, second and third positions are the same, it means that there are multiple data to be identified at the first, second or third position;

[0133] Arrange the multiple data to be identified that are located at the first, second or third position in descending order according to the remark value from large to small, and obtain the data arrangement row in descending order; since the remark value corresponds to the number of remark words that are the same as the search feature in the data to be identified, the larger the remark value, the more remark words that are the same as the search feature in the data to be identified, and the higher the matching degree between the data to be identified and the search feature, so that the multiple data to be identified that have the same matching value can be accurately identified again;

[0134] The data to be identified at the first, second and third positions in the descending data arrangement row are recorded as target data.

[0135] It should be noted that by marking the top three matching data to be identified as target data, it can be ensured that the stored data corresponding to the target data can be truly and accurately matched with the requested data, thereby providing a data basis for subsequent execution operations.

[0136] S5: formulating data interaction instructions according to the execution characteristics, identifying the priority of the data interaction instructions, and controlling the target data to execute the data interaction instructions on the TV side;

[0137] Data interaction instructions refer to instructions that can perform specific operations on target data, which are used to represent different execution features with different instructions and serve as specific measures for subsequent execution of target data, thereby achieving effective execution operations on target data;

[0138] The data interaction instructions include form determination instructions, mode selection instructions and time jump instructions; the form determination instructions are used to represent the structure form of the target data, and are used to implement different form determination operations; the mode selection instructions are used to represent the mode of the target data, and are used to implement different mode selection operations; the time jump instructions are used to represent the time of the target data, and are used to implement different time jump operations;

[0139] The formulation method of the form determination instruction, the mode selection instruction and the time jump instruction includes:

[0140] Identify the second keywords of the execution features one by one through natural language processing technology to obtain m second keywords;

[0141] Record the form structure, form specification and form attribute as the first instruction word;

[0142] Recording the working mode, mode type and mode attribute as the second instruction word;

[0143] Record time jump, time switch and time attribute as the third instruction word;

[0144] When the first instruction word exists in the m second keywords, it indicates that a determination operation needs to be performed on the form of the target data, and a form determination instruction is formulated;

[0145] When the second instruction word exists in the m second keywords, it indicates that a mode selection operation needs to be performed on the target data, and a mode selection instruction is formulated;

[0146] When the third instruction word exists in the m second keywords, it means that a time jump operation needs to be performed on the target data, and a time jump instruction is formulated.

[0147] Once the data interaction instructions are formulated, the target data can be executed through the data interaction instructions, so that the target data can be displayed on the TV end and meet the data request requirements of the remote control end, thereby achieving the effect of AI question-answering interaction;

[0148] When there are more than one data interaction instructions, multiple data interaction instructions need to be executed in sequence to prevent mutual interference when multiple data interaction instructions are executed synchronously, thereby ensuring that the data interaction instructions can perform efficient and orderly execution operations on the target data. Therefore, it is necessary to determine the priority of the data interaction instructions;

[0149] Since the form determination instruction is directly used to operate the form structure of the target data, it can have a decisive impact on the type and form of the target data. Therefore, the form determination instruction has the highest priority, while the time jump instruction is used to represent the time point of the target data during execution and will not have a substantial impact on the target data itself. Therefore, the time jump instruction has the lowest priority;

[0150] In summary, the priority is: the priority of the form determination instruction is higher than the priority of the mode selection instruction, and the priority of the mode selection instruction is higher than the priority of the time jump instruction;

[0151] The TV end refers to the port that can output the data matched and retrieved by the AI ​​server. As the output end of AI question and answer interaction, when the priority of the data interaction instruction is identified, the formulated data interaction instruction can be used to perform relevant execution operations on the target data in sequence according to the priority. In this way, question and answer interaction processing can be performed from three dimensions: question and answer form, question and answer mode, and question and answer time of the target data, thereby ensuring efficient and accurate AI question and answer interaction effect between the remote control end and the TV end.

[0152] In this embodiment, by receiving the request data from the remote control end, and based on the feature extraction criteria, extracting the request features from the request data, based on the compliance criteria, performing compliance identification on the request features, identifying the compliance status of the request features, and determining whether to enter the question-and-answer interaction mode, identifying the feature attributes of the request features one by one, and dividing the request features into retrieval features and execution features according to the feature attributes, filtering out the data to be identified that matches the retrieval features from the AI ​​database, calculating the matching degree of the data to be identified, marking the target data from the data to be identified, formulating the data interaction instructions according to the execution features, identifying the priority of the data interaction instructions, and controlling the target data on the TV end to execute the data interaction instructions; compared with the prior art, on the basis of the feature extraction criteria, It is capable of extracting request features from received request data, and identifying the validity of request features to see whether they meet the requirements of AI question-and-answer interaction, thereby identifying illegal and unreasonable AI question-and-answer interaction requests at the source, which not only raises the threshold for AI question-and-answer interaction, but also reduces the workload of AI question-and-answer interaction. At the same time, it divides request features into retrieval features and execution features to ensure that the AI ​​server can match target data based on retrieval features, and execute target data in an orderly manner through the TV end based on the execution features, thereby ensuring that target data corresponding to the request data can be executed accurately and orderly, thereby improving the fluency and orderliness of AI question-and-answer interaction, and meeting people's high-quality requirements for AI question-and-answer interaction through the TV voice remote control.

[0153] Example 2: Please refer to Figure 2 As shown, the part not described in detail in this embodiment is described in the first embodiment, and a TV voice remote control AI question-answering interactive system is provided, which is applied to an AI server and is used to implement a TV voice remote control AI question-answering interactive method, including a feature extraction module, a recognition and determination module, a feature division module, a matching recognition module and an instruction execution module, wherein each module is connected via a wired or wireless network;

[0154] The feature extraction module is used to receive the request data from the remote control end, and extract the request feature from the request data based on the feature extraction criterion. The feature extraction criterion is: take the first character as the starting position and the last character as the ending position, and perform feature extraction in sequence;

[0155] An identification and determination module is used to identify the compliance of request features based on compliance criteria, identify the compliance status of the request features, and determine whether to enter the question-and-answer interaction mode. The compliance status includes effective compliance and invalid compliance.

[0156] A feature division module is used to identify the feature attributes of the request features one by one, the feature attributes include noun attributes and verb attributes, and divide the request features into retrieval features and execution features according to the feature attributes;

[0157] The matching recognition module is used to filter out the data to be recognized that matches the search features from the AI ​​database, calculate the matching degree of the data to be recognized, and mark the target data from the data to be recognized;

[0158] The instruction execution module is used to formulate data interaction instructions according to the execution characteristics, identify the priority of the data interaction instructions, and control the target data to execute the data interaction instructions on the TV side.

[0159] The above description is only a specific implementation mode of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. An AI question-answering interactive method based on a TV voice remote control, applied to an AI server, characterized in that: include: S1: receiving request data from the remote control end, and extracting request features from the request data based on a feature extraction criterion, wherein the feature extraction criterion is: taking the first character as the start position and the last character as the end position, and performing feature extraction in sequence; S2: Based on the compliance criteria, the request features are identified for compliance, the compliance status of the request features is identified, and it is determined whether to enter the question-and-answer interaction mode. The compliance status includes valid compliance and invalid compliance. If the question-and-answer interaction mode is entered, S3 is executed; if the question-and-answer interaction mode is not entered, S1-S2 are repeated. S3: identifying the feature attributes of the request features one by one, the feature attributes include noun attributes and verb attributes, and dividing the request features into retrieval features and execution features according to the feature attributes; S4: Filter out the data to be identified that matches the search feature from the AI ​​database, calculate the matching degree of the data to be identified, and mark the target data from the data to be identified; S5: Formulate data interaction instructions according to the execution characteristics, identify the priority of the data interaction instructions, and control the target data on the TV side to execute the data interaction instructions.

2. The AI ​​question-answering interactive method based on a TV voice remote controller according to claim 1, characterized in that: The method for extracting the request feature includes: Receive the voice information of the personnel in real time through the remote control terminal, and record the time of first receiving the voice information as the starting time; When no more voice information is received within the preset receiving time, the time when the voice information is last received is recorded as the end time, and the voice information received during the period from the start time to the end time is recorded as the request data; Mark all characters in the request data one by one, and arrange all characters in order according to the time of receipt to generate a character row; Taking the first character in the character row as the starting point, the key characters in the character row are identified one by one through natural language processing technology, and the association status of the key characters in adjacent positions is identified through the voice model; Taking the last character in the character row as the end point, i key characters with associated status are combined head to tail to obtain i request features.

3. The AI ​​question-answering interactive method based on a TV voice remote controller according to claim 2, characterized in that: The compliance criteria are: the number of abnormal characters in the compliant request feature is less than or equal to 1; Methods for identifying effective and ineffective compliance include: Mark the characters of i requested features one by one, and compare the characters of i requested features with the compliant characters in the compliant database one by one for consistency; Characters that are inconsistent with the compliant characters in the compliant database are recorded as abnormal characters, and the number of abnormal characters in i request features is counted one by one to obtain i abnormal values; The compliance status of the request feature with an outlier value less than or equal to 1 is recorded as valid compliance; The compliance status of the request feature with an outlier value greater than 1 is recorded as invalid compliance.

4. The AI ​​question-answering interactive method based on a TV voice remote controller according to claim 3, characterized in that: The method for determining whether to enter the question-answer interaction mode includes: Counting the number of request features corresponding to effective compliance and the number of request features corresponding to invalid compliance one by one, and obtaining a first compliance value and a second compliance value respectively; comparing the first compliance value and the second compliance value to generate a compliance ratio; The expression of the compliance ratio is: In the formula, HG bz is the compliance ratio, HG d1 is the first compliance value, HG d2 is the second compliance value; The compliance ratio HG bz Compared with the preset compliance threshold HG yz Compare; When HG bz Greater than or equal to HG yz When , it is determined to enter the question-answer interaction mode; When HG bz Smaller than HG yz , it is determined not to enter the question-and-answer interactive mode.

5. The AI ​​question-answering interactive method based on a TV voice remote controller according to claim 4, characterized in that: The method for identifying noun attributes and verb attributes includes: Arrange the i request features in chronological order, and repeatedly compare the characters of the i request features with the words in the noun database and the verb database in turn; When the characters of the requested feature are repeated with words in the noun database, the feature attribute of the requested feature is identified as a noun attribute; When characters of the request feature are repeated with words in the verb database, the feature attribute of the request feature is identified as a verb attribute.

6. The AI ​​question-answering interactive method based on a TV voice remote controller according to claim 5, characterized in that: The method for dividing the retrieval features and the execution features includes: When the feature attribute of the request feature is a noun attribute, the request feature is divided into retrieval features to obtain p retrieval features; When the feature attribute of the request feature is a verb attribute, the request feature is divided into execution features to obtain m execution features.

7. The AI ​​question-answering interactive method based on a TV voice remote controller according to claim 6, characterized in that: The calculation method of the matching degree includes: Using natural language processing technology, the first keywords of w stored data in the AI ​​database are identified one by one to obtain w first keywords; The first keyword repeated with any search feature is recorded as the target keyword, and the stored data corresponding to the target keyword is recorded as the data to be identified, and X data to be identified are obtained; Through the attribute management system, X attribute boxes of the data to be identified are searched one by one, the remark words in the attribute boxes are marked, and the number of the remark words is counted to obtain X total values; The remark words that are the same as the search features are recorded as valid remark, and the number of valid remark in the X data to be identified is counted to obtain X remark values; Compare X remark values ​​with X total values ​​one by one to obtain X matching degrees; The expression of matching degree is: In the formula, PD dx is the matching degree of the xth data to be identified, BZ zx is the remark value of the xth data to be identified, ZL zx is the total value of the xth data to be identified.

8. The AI ​​question-answering interactive method based on a TV voice remote controller according to claim 7, characterized in that: The target data marking method includes: Arrange the X pieces of data to be identified in order from largest to smallest matching degree; When the matching degrees of the first, second and third to-be-identified data are all inconsistent, the first, second and third to-be-identified data are recorded as target data; When the matching degrees of the first, second and third data to be identified are the same, the multiple data to be identified that are located in the first, second or third positions are arranged in descending order according to the remark values ​​from large to small to obtain the data arrangement rows in descending order; The data to be identified at the first, second and third positions in the descending data arrangement row are recorded as target data.

9. The AI ​​question-answering interactive method based on a TV voice remote controller according to claim 8, characterized in that: The data interaction instructions include form determination instructions, mode selection instructions and time jump instructions; The formulation method of the form determination instruction, the mode selection instruction and the time jump instruction includes: Identify the second keywords of the execution features one by one through natural language processing technology to obtain m second keywords; Record the form structure, form specification and form attribute as the first instruction word; Recording the working mode, mode type and mode attribute as the second instruction word; Record time jump, time switch and time attribute as the third instruction word; When the first instruction word exists in the m second keywords, a form determination instruction is formulated; When the second instruction word exists in the m second keywords, a mode selection instruction is formulated; When the third instruction word exists among the m second keywords, a time jump instruction is formulated.

10. The AI ​​question-answering interactive method based on a TV voice remote controller according to claim 9, characterized in that: The priorities are: the priority of the form determination instruction is higher than the priority of the mode selection instruction, and the priority of the mode selection instruction is higher than the priority of the time jump instruction.

Images