Intelligent robot interaction method and system based on deep learning

Abstract

The invention relates to the technical field of robot interaction, and discloses an intelligent robot interaction method and system based on deep learning, and the system comprises an interaction instruction analysis module, an instruction processing module, and a multi-user management module. Known and unknown instructions are classified accurately and efficiently; known instructions are quickly executed according to a scheme with the maximum similarity, keywords are firstly extracted from unknown instructions, related processing schemes are retrieved, the keywords are recombined when no adaptive scheme exists, information is searched and screened by means of a knowledge graph to generate a temporary scheme, and the temporary scheme is confirmed by a user, so that the processing capacity of the unknown instructions is enhanced; user identities are determined through biological recognition, priorities are set according to the user identities and task emergency degrees, instructions are arranged according to the priorities and processed, and results are fed back after execution, so that the problem of multi-user interaction disorder is solved, interaction management and task processing efficiency is improved, system intelligence and practicability are improved, and user interaction experience is optimized. And the competitiveness of the robot in practical application is enhanced.

Description

Technical Field

[0001] The present invention relates to the field of robot interaction technology, and in particular to an intelligent robot interaction method and system based on deep learning. Background Art

[0002] With the rapid development of artificial intelligence (AI), intelligent robots are increasingly being used in a wide range of fields, including home, healthcare, education, and services. The performance of intelligent robot interaction systems, a crucial bridge between humans and robots, directly impacts the user experience and the practical application value of robots.

[0003] Currently, intelligent robot interaction systems face numerous challenges in command parsing and processing. For one thing, user commands are diverse and flexibly expressed. Traditional command parsing methods often rely on fixed matching rules, making it difficult to accurately understand commands with ambiguous semantics or poorly expressed specifications. Furthermore, they are unable to effectively process unknown commands that have not been entered into the system. This limits the robot's interactive capabilities and makes it difficult to meet the diverse needs of users. Furthermore, existing methods for calculating command similarity are either computationally complex and inefficient, making them difficult to run quickly on resource-limited robotic devices; or they are overly simplistic, unable to accurately measure the semantic similarity between commands, affecting the accuracy of command classification.

[0004] The problem is even more pronounced in multi-user interaction scenarios. When multiple users simultaneously send interaction requests to the robot, existing systems lack effective multi-user management mechanisms and are unable to scientifically determine user priorities. This can easily lead to chaotic command processing and delayed responses to important and urgent tasks, significantly reducing the robot's interaction efficiency and service quality in multi-user environments.

[0005] To sum up, developing an intelligent robot interaction system that can accurately parse various user commands, efficiently process unknown commands, and scientifically manage multi-user interactions has become the key to improving the practicality and competitiveness of intelligent robots and needs to be solved urgently. Summary of the Invention

[0006] The purpose of the present invention is to provide an intelligent robot interaction method and system based on deep learning, which solves the technical problems raised in the background technology.

[0007] The purpose of the present invention can be achieved through the following technical solutions:

[0008] An intelligent robot interaction system based on deep learning, comprising:

[0009] Interactive instruction parsing module, used to parse and classify the instructions input by the user;

[0010] The parsing classification method is: traverse the pre-set instruction database, each instruction Di , i = 1, 2, ... n, n is the total number of instructions in the instruction database;

[0011] Extract the instruction D0 input by the user; then compare the instruction D0 input by the user with each instruction D in the instruction database. i Perform similarity calculation and compare the obtained similarity S with the preset similarity threshold T to determine the instruction type of the user input instruction;

[0012] The instruction processing module is used to process the instructions input by the user according to the analysis and classification results of the interactive instruction parsing module;

[0013] The multi-user management module is used to perform multi-user interaction management analysis when the intelligent robot faces interaction requests from multiple users at the same time, and respond to the interaction requests of multiple users one by one based on the results of the multi-user interaction management analysis.

[0014] As a further solution of the present invention: the similarity calculation method adopts the edit distance algorithm in the text similarity calculation method, which is as follows:

[0015] Among them, the edit distance refers to the minimum number of editing operations required to transform one string into another;

[0016] Extract the character length M0 of the user input instruction D0 and each instruction D in the instruction database i The character length M i ;

[0017] Construct a (M0+1)×(M i +1) matrix E;

[0018] Right now:

[0019] Among them, E(uv i ) means the first u characters of D0 and D i The front v i The edit distance between characters, and 1≤u≤M0, 1≤v i ≤M i ;

[0020] When E(00)=0, it means the edit distance between two empty strings is 0;

[0021] For u = 1 to M0, E(u0) = u, which means that it takes u deletion operations to transform the first u characters of D0 into an empty string;

[0022] For v i =1 to M i ,E(0v i )=v i, which means transforming the empty string into D i The front v i characters need v i Insert operation;

[0023] For u=1 to M0, v i =1 to M i , if the uth character of D0 is the same as D i No. v i characters are the same, then:

[0024] E(uv i )=min{E(u-1,v i -1),E(u-1,v i )+1,E(u,v i -1)+1};

[0025] Among them, E(u-1,j-1) means that the current characters are the same and no additional operation is required. E(u-1,v i )+1 means deleting the uth character of D0, E(u,v i -1)+1 means insert D into I i No. v i characters;

[0026] If the uth character of D0 is the same as D i No. v i characters are different, then:

[0027] E(uv i )=min{E(u-1,v i -1)+1,E(u-1,v i )+1,E(u,v i -1)+1};

[0028] Among them, E(u-1,v i -1)+1 means replace the uth character of D0 with D i No. v i characters;

[0029] pass Calculate the difference between the user input command D0 and the command D in the command database i The similarity S(D0, D i );

[0030] When S(D0, D i )≥T, then the instruction D0 is determined to be a known instruction; otherwise, the instruction D0 is determined to be an unknown instruction.

[0031] As a further solution of the present invention, the similarity calculation method also adopts a simple similarity calculation method based on character matching, which is as follows:

[0032] First, extract the character length M0 of the user input instruction D0 and each instruction D in the instruction database i The character length M i ;

[0033] Then define a counter C=0 to record the number of matching characters;

[0034] From D0 and D i Comparison starts with the first character:

[0035] When comparing to the e-th character, if the e-th character of D0 is the same as D i The e-th character is the same as that of , then C=C+1;

[0036] Where, e=1, ...min(M0,M i );

[0037] After completing the comparison, pass: Calculate the difference between the user input command D0 and the command D in the command database i The similarity S(D0, D i ).

[0038] As a further solution of the present invention: the instruction processing method is as follows:

[0039] When it is determined that instruction D0 is a known instruction, the instruction input by the user is executed according to the execution scheme of the known instruction with the greatest similarity to instruction D0;

[0040] When it is determined that the instruction D0 is an unknown instruction, a depth matching process is performed on the instruction D0, and then an execution plan of the user input instruction is determined based on the depth matching process.

[0041] As a further solution of the present invention: the depth matching processing method is as follows:

[0042] The keyword extraction method based on word frequency statistics is used to obtain the keyword set K of D0 in the unknown instruction = {K z}, where z = 1, 2, ... t, t is the number of keywords;

[0043] In the pre-set extended solution database, the solution related to the keyword set K is retrieved, and each solution in the extended solution database is recorded as F j , j = 1, 2, ... m, m is the total number of records of treatment solutions in the extended treatment solution database DB;

[0044] Then, D0 in the unknown instruction and each processing solution F j The matching degree is calculated as follows:

[0045] Extract each treatment plan F j The keyword set K contained in j ;

[0046] Calculate K and K first j The number of intersection elements Na, that is, Na=|K∩K j ∣;

[0047] Then calculate K and K j The number of elements in the union of Nb, that is, Nb = |K∪K j ∣;

[0048] Then through: Calculate D0 and each processing solution F in the unknown instruction j The matching degree M(D0,F j );

[0049] Then D0 in the unknown instruction and each processing solution F j The matching degree M(D0,F j ) is compared with the pre-set matching threshold My:

[0050] If there exists a condition that satisfies M(D0,F j )≥My, then select the corresponding solution F j to handle the unknown instruction.

[0051] As a further solution of the present invention: If there is no j )≥My's j, that is, there is no adapted solution in the extended solution database, the robot will execute the following strategies, as follows:

[0052] Step 1: Recombine and expand the keyword set K by arranging and combining the keywords in different orders to generate a new instruction description set C = {C g}, where g = 1, 2, ..., s, where s is the number of instruction descriptions in set C;

[0053] Step 2: For each instruction description C in the instruction description set C g , perform associative search in pre-established general knowledge graph;

[0054] Among them, the general knowledge graph stores a large number of semantic associations between words and concepts, and searches for C g Related processing suggestions or information, forming a related information set R g ;

[0055] Step 3: From R g The most relevant information is selected and the relevance score is based on the number of keyword matches.

[0056] Count the number of keywords that match the keyword set K in each information node, and use it to calculate the number of keywords in R g Sort all the information nodes in the , and then give priority to the information nodes with high ranking and high score, and get a temporary solution for the unknown instruction;

[0057] Step 4: Show the user a temporary solution and ask, "Do you want to implement this solution?"

[0058] If the user confirms, the temporary solution will be implemented;

[0059] If the user denies the request, the system will request the user to provide more instruction details or related information, and then re-parse, classify and process the instructions input by the user.

[0060] As a further solution of the present invention: the relevant information set R g The specific formation method is as follows:

[0061] During the search, first parse C g Keywords and key concepts in

[0062] Based on the pre-built relationship paths in the knowledge graph, starting from these keywords, traverse along various relationship edges;

[0063] The robot will follow these relationship edges and gradually expand the search range to find g Related processing suggestions or information, forming a related information set R g .

[0064] As a further solution of the present invention: the multi-user interaction management and analysis method is as follows:

[0065] Step G1: First, user identification is performed. Specifically, the identity information of each user is determined by biometric identification means corresponding to face recognition technology and voice recognition technology, and the user identity information set is formed as U = {U f}, f = 1, 2, ... h, h is the number of users;

[0066] Step G2: After identifying the user identity, the user priority is determined according to a pre-set priority classification rule;

[0067] Step G3: Arrange the commands input by each user into a task queue according to the user priority from high to low.

[0068] In the task queue, the interactive instruction parsing module first parses the highest priority user input instruction in the task queue to determine whether it is a known instruction or an unknown instruction;

[0069] If the command is known, the command input by the user is directly executed according to the corresponding execution plan;

[0070] If it is an unknown instruction, the instruction processing module performs unknown instruction processing on the instruction processing method for the unknown instruction;

[0071] After the instruction is executed, the completed instruction is removed from the task queue and the next highest priority task is processed until the task queue is empty.

[0072] As a further solution of the present invention: wherein, the priority division rules include:

[0073] Different users are assigned different basic priority values A1 according to their identities;

[0074] Where A1={A1 x1}, x1 = 1, 2, ..., q1, q1 refers to the number of attributes corresponding to the user identity;

[0075] Different users are assigned different emergency priority values A2 according to the urgency of the task corresponding to the user input instruction;

[0076] Where A2={A2 x2}, x2=1, 2, ... q2, q2 refers to the number of task types corresponding to the urgency of the task.

[0077] As a further solution of the present invention: the user priority is determined as follows:

[0078] Extract the basic priority value A1 corresponding to each user f and emergency priority value A2 f ;

[0079] Then passed: P f =A1 f ×α1+A2 f ×α2, calculate the user priority P f ;

[0080] Where α1 and α2 are preset weight coefficients corresponding to the basic priority value and the emergency priority value.

[0081] A deep learning-based intelligent robot interaction method, implemented by a deep learning-based intelligent robot interaction system, comprises the following steps:

[0082] Step 1: Interactive instruction analysis:

[0083] Parse and classify the instructions input by the user to determine the instruction type of the user input instruction;

[0084] Step 2: Single instruction processing:

[0085] The interactive instruction parsing module analyzes and classifies the user input instructions, processes the instructions, and determines the execution plan of the intelligent robot;

[0086] Step 3: Multi-command management:

[0087] When the intelligent robot faces interaction requests from multiple users at the same time, it conducts multi-user interaction management analysis to respond to the interaction requests of multiple users one by one.

[0088] Beneficial effects of the present invention:

[0089] At the instruction parsing level, the interactive instruction parsing module of the present invention adopts an edit distance algorithm and a simple similarity calculation method based on character matching. The former can accurately measure the similarity between user input instructions and database instructions by quantifying the number of editing operations of string transformation, and effectively identify complex instructions; the latter has concise and efficient calculation logic and can quickly process instructions. The combination of the two provides dual guarantees of flexibility and accuracy for instruction parsing, ensuring that all types of user instructions can be accurately classified as known or unknown instructions.

[0090] In the instruction processing link of the present invention, for known instructions, the system directly calls the execution plan of the instruction with the highest similarity to achieve rapid response; when facing unknown instructions, it first uses word frequency statistics to extract keywords, and searches for matches in the extended processing solution database. When there is no adapted solution, it reorganizes and expands through keywords, uses the general knowledge graph to perform semantic association search, filters the most relevant information to generate a temporary processing solution, and confirms with the user, which greatly improves the system's processing ability and adaptability to unknown instructions, allowing the robot to cope with complex and diverse user needs.

[0091] In terms of multi-user management, the present invention uses a multi-user management module that accurately identifies user identities through biometric technologies such as face recognition and voice recognition, sets priority division rules based on user identities and task urgency, scientifically calculates user priorities, arranges user instructions into task queues according to priority, gives priority to high-priority instructions, ensures timely response to urgent and important tasks, and feeds back results after the instructions are executed, effectively solving the confusion problem when multiple users interact at the same time, significantly improving the robot's interaction management and task processing efficiency in multi-user scenarios, and optimizing the multi-user usage experience.

[0092] Overall, the intelligent robot interaction system of the present invention realizes the accurate recognition and efficient processing of user commands and orderly interaction in multi-user scenarios through the collaborative work of multiple modules such as command parsing, processing and multi-user management, which significantly enhances the intelligence, practicality and user interaction satisfaction of the system, and greatly improves the interactive performance and competitiveness of intelligent robots in practical applications. BRIEF DESCRIPTION OF THE DRAWINGS

[0093] The present invention will be further described below with reference to the accompanying drawings.

[0094] Figure 1 This is a system block diagram of an intelligent robot interaction system based on deep learning in the present invention.

[0095] Figure 2 This is a flow chart of an intelligent robot interaction method based on deep learning in the present invention. DETAILED DESCRIPTION

[0096] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. 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 making any creative efforts shall fall within the scope of protection of the present invention.

[0097] As embodiment 1 of the present invention:

[0098] See also Figure 1 and Figure 2 As shown, the present invention is an intelligent robot interaction system based on deep learning, comprising:

[0099] Interactive instruction parsing module, used to parse and classify the instructions input by the user;

[0100] The specific method is as follows:

[0101] In the pre-set instruction database, traverse each instruction D i , i = 1, 2, ... n, n is the total number of instructions in the instruction database;

[0102] Extract the instruction D0 input by the user;

[0103] Then the user input command D0 is compared with each command D in the command database. i Perform similarity calculation and compare the obtained similarity S with the preset similarity threshold T to determine the instruction type of the user input instruction;

[0104] The similarity calculation method uses the edit distance algorithm in the text similarity calculation method, as follows:

[0105] Among them, the edit distance refers to the minimum number of editing operations required to transform one string into another;

[0106] Extract the character length M0 of the user input instruction D0 and each instruction D in the instruction database i The character length M i ;

[0107] Construct a (M0+1)×(M i +1) matrix E;

[0108] Right now:

[0109] Among them, E(uv i ) means the first u characters of D0 and D i The front v i The edit distance between characters, and 1≤u≤M0, 1≤v i ≤M i ;

[0110] When E(00)=0, it means the edit distance between two empty strings is 0;

[0111] For u = 1 to M0, E(u0) = u, which means that it takes u deletion operations to transform the first u characters of D0 into an empty string;

[0112] For v i =1 to M i ,E(0v i )=v i , which means transforming the empty string into D i The front v i characters need v i Insert operation;

[0113] For u=1 to M0, v i =1 to M i , if the uth character of D0 is the same as D i No. v i characters are the same, then:

[0114] E(uv i )=min{E(u-1,v i -1),E(u-1,v i )+1,E(u,v i -1)+1};

[0115] Among them, E(u-1,j-1) means that the current characters are the same and no additional operation is required. E(u-1,v i)+1 means deleting the uth character of D0, E(u,v i -1)+1 means insert D into I i No. v i characters;

[0116] If the uth character of D0 is the same as D i No. v i characters are different, then:

[0117] E(uv i )=min{E(u-1,v i -1)+1,E(u-1,v i )+1,E(u,v i -1)+1};

[0118] Among them, E(u-1,v i -1)+1 means replace the uth character of D0 with D i No. v i characters;

[0119] pass Calculate the difference between the user input command D0 and the command D in the command database i The similarity S(D0, D i );

[0120] Among them, the similarity value range is between 0 and 1, and the larger the value, the more similar the two are;

[0121] When S(D0, D i )≥T, then the instruction D0 is determined to be a known instruction; otherwise, the instruction D0 is determined to be an unknown instruction;

[0122] The instruction processing module is used to process the instructions input by the user according to the analysis and classification results of the interactive instruction parsing module;

[0123] The instructions are processed as follows:

[0124] When it is determined that instruction D0 is a known instruction, the instruction input by the user is executed according to the execution scheme of the known instruction with the greatest similarity to instruction D0;

[0125] When the instruction D0 is determined to be an unknown instruction, a keyword extraction method based on word frequency statistics is used to obtain the keyword set K of D0 in the unknown instruction = {K z}, where z = 1, 2, ... t, t is the number of keywords;

[0126] For example, for the instruction “find a place to watch a movie”, we can extract keywords such as “watch a movie” and “place” through word frequency statistics;

[0127] In the pre-set extended solution database, the solution related to the keyword set K is retrieved, and each solution in the extended solution database is recorded as F j , j = 1, 2, ... m, m is the total number of records of treatment solutions in the extended treatment solution database DB;

[0128] Then, D0 in the unknown instruction and each processing solution F j The matching degree is calculated as follows:

[0129] Extract each treatment plan F j The keyword set K contained in j ;

[0130] Calculate K and K first j The number of intersection elements Na, that is, Na=|K∩K j ∣;

[0131] Then calculate K and K j The number of elements in the union of Nb, that is, Nb = |K∪K j ∣;

[0132] Then through: Calculate D0 and each processing solution F in the unknown instruction j The matching degree M(D0,F j );

[0133] Among them, the matching degree M(D0,F j ) ranges from 0 to 1. A larger value indicates a higher degree of matching between instruction Y and solution yi.

[0134] Then D0 in the unknown instruction and each processing solution F j The matching degree M(D0,F j ) is compared with the pre-set matching threshold My:

[0135] In this embodiment, My is set to 0.6, which can be flexibly adjusted according to the actual application scenario;

[0136] If there exists a condition that satisfies M(D0,F j )≥My, then select the corresponding solution F j To process the unknown instruction;

[0137] If there is no such condition that M(D0,F j )≥My's j, that is, there is no adapted solution in the extended solution database, the robot will execute the following strategies, as follows:

[0138] Step 1: Recombine and expand the keyword set K by arranging and combining the keywords in different orders to generate a new instruction description set C = {C g}, where g = 1, 2, ..., s, where s is the number of instruction descriptions in set C;

[0139] For example, if the keyword set is book purchase, then set C may contain instruction descriptions such as "buy books" and "book purchase";

[0140] Step 2: For each instruction description C in the instruction description set C g , perform associative search in pre-established general knowledge graph;

[0141] Among them, the general knowledge graph stores a large number of semantic associations between words and concepts, and searches for C g Related processing suggestions or information, forming a related information set R g ;

[0142] Related information set R g The specific formation method is as follows:

[0143] During the search, first parse C g Keywords and key concepts in

[0144] For example, for the situation description “buy red apples”, the robot will extract keywords such as “buy”, “red”, and “apple”;

[0145] Based on the pre-built relationship paths in the knowledge graph, starting from these keywords, traverse along various relationship edges;

[0146] For example, in the knowledge graph, the entity "apple" has a category relationship with "fruit category", an attribute relationship with "apple origin", and a semantic association relationship with "channels for purchasing apples";

[0147] The robot will follow these relationship edges and gradually expand the search range to find g Related processing suggestions or information, forming a related information set R g ;

[0148] For example, during the traversal process, all relevant information nodes are collected. These nodes contain various possible information, such as purchase recommendations for brightly colored and smooth-skinned apples, addresses of nearby fruit stores, and purchase channel information corresponding to links to online e-commerce platforms.

[0149] Step 3: From R g The most relevant information is selected and the relevance score is based on the number of keyword matches.

[0150] Count the number of keywords that match the keyword set K in each information node, and use it to calculate the number of keywords in R g Sort all the information nodes in the , and then give priority to the information nodes with high ranking and high score, and get a temporary solution for the unknown instruction;

[0151] Among them, the greater the number of matches, the higher the relevance score;

[0152] For example, if an information node mentions "fresh red apples can be purchased at a nearby fruit store," where the three keywords "purchase," "red," and "apple" all match K, the node will have a higher relevance score than an information node that only matches one keyword.

[0153] Step 4: Show the user a temporary solution and ask, "Do you want to implement this solution?"

[0154] If the user confirms, the temporary solution will be implemented;

[0155] If the user denies the request, the system will request the user to provide more instruction details or related information, and then re-parse, classify and process the instructions input by the user.

[0156] In the first embodiment, the interactive instruction parsing module uses the edit distance algorithm to calculate the similarity between the user input instruction and the instruction in the instruction database, which can accurately parse the user instruction and distinguish between known instructions and unknown instructions. For known instructions, they can be quickly executed according to the corresponding execution plan; for unknown instructions, by extracting keywords based on word frequency statistics, searching for relevant processing solutions in the extended processing solution database, and when there is no adaptation solution, using the general knowledge graph to perform an associated search to generate a temporary processing solution, the intelligent robot's processing ability and adaptability to various user instructions are greatly improved. When faced with complex and diverse user instructions, whether they are known or unknown, the robot can give as effective a processing method as possible, thereby improving the practicality and intelligence of the interactive system and enhancing the fluency and satisfaction of the user's interaction with the intelligent robot.

[0157] As the second embodiment of the present invention:

[0158] See also Figure 1 and Figure 2 As shown, in the specific implementation of this application, compared with Example 1, the technical solution of this embodiment is different from that of Example 1 only in that in this embodiment, the similarity calculation method can also adopt a simple similarity calculation method based on character matching, which is as follows:

[0159] First, extract the character length M0 of the user input instruction D0 and each instruction D in the instruction database i The character length Mi ;

[0160] Then define a counter C=0 to record the number of matching characters;

[0161] From D0 and D i Comparison starts with the first character:

[0162] When comparing to the e-th character, if the e-th character of D0 is the same as D i The e-th character is the same as that of , then C=C+1;

[0163] Where, e=1, ...min(M0,M i );

[0164] After completing the comparison, pass: Calculate the difference between the user input command D0 and the command D in the command database i The similarity S(D0, D i ).

[0165] Example 2 uses a simple similarity calculation method based on character matching. This method has simple calculation logic and low computational complexity when calculating the similarity between user input instructions and instructions in the instruction database. It can quickly complete the calculation process, reduce the consumption of system computing resources, and improve the efficiency of instruction parsing. This simple calculation method is suitable for scenarios with high computing performance requirements or small instruction databases. It provides a more lightweight instruction similarity calculation option for intelligent robot interaction systems, allowing the system to flexibly select the appropriate calculation method based on actual needs in different application environments, thereby enhancing the applicability and flexibility of the system.

[0166] As the third embodiment of the present invention:

[0167] See also Figure 1 and Figure 2 As shown, in the specific implementation of this application, compared with Example 1 and Example 2, the technical solution of this embodiment is to combine the solutions of Example 1 and Example 2. The technical solution of this embodiment differs from that of Example 1 and Example 2 only in that this embodiment further includes:

[0168] The multi-user management module is used to perform multi-user interaction management and analysis when the intelligent robot faces interaction requests from multiple users at the same time. The specific methods are as follows:

[0169] Step G1: First, user identification is performed. Specifically, the identity information of each user is determined by biometric identification means corresponding to face recognition technology and voice recognition technology, and the user identity information set is formed as U = {U f}, f = 1, 2, ... h, h is the number of users;

[0170] In this embodiment, the face recognition technology and the voice recognition technology are existing technologies and therefore will not be described in detail.

[0171] Step G2: After identifying the user identity, determine the user priority according to the pre-set priority classification rules;

[0172] The priority classification rules include:

[0173] Different users are assigned different basic priority values A1 according to their identities;

[0174] Where A1={A1 x1}, x1 = 1, 2, ..., q1, q1 refers to the number of attributes corresponding to the user identity;

[0175] In this embodiment, for example, the basic priority coefficients of the elderly are A11 = 0.8, those of children are A12 = 0.7, those of ordinary family members are A13 = 0.5, and those of visitors are A14 = 0.3;

[0176] Different users are assigned different emergency priority values A2 according to the urgency of the task corresponding to the user input instruction;

[0177] Where A2={A2 x2}, x2 = 1, 2, ..., q2, q2 refers to the number of task types corresponding to the urgency of the task;

[0178] In this embodiment, the task types include safety tasks, daily life tasks, and leisure and entertainment tasks. For example, the safety tasks "There is smoke at home" and "Someone fell down" are assigned an emergency priority value A21=0.9; the daily life tasks "Help get medicine" and "Urgent need drinking water" are assigned an emergency priority value A21=0.7; and the leisure and entertainment tasks "Play music" and "Tell a joke" are assigned an emergency priority value A21=0.3.

[0179] User priority is determined as follows:

[0180] Extract the basic priority value A1 corresponding to each user f and emergency priority value A2 f ;

[0181] Then passed: P f =A1 f ×α1+A2 f ×α2, calculate the user priority P f ;

[0182] Where α1 and α2 are the preset weight coefficients corresponding to the basic priority value and the emergency priority value;

[0183] Step G3: According to the user priority P of each user f ,The commands input by each user are arranged into a task queue according to the user priority from high to low;

[0184] In the task queue, the interactive instruction parsing module first parses the highest priority user input instruction in the task queue to determine whether it is a known instruction or an unknown instruction;

[0185] If the command is known, the command input by the user is directly executed according to the corresponding execution plan;

[0186] If it is an unknown instruction, the instruction processing module performs unknown instruction processing on the instruction processing method for the unknown instruction;

[0187] After the instruction is executed, the completed instruction is removed from the task queue and the next highest priority task is processed until the task queue is empty.

[0188] In this embodiment, after the instruction is executed, the execution result is also fed back to the user who input the instruction, such as "relevant medical information has been found for you" or "sorry, I did not understand your instruction, please explain it again."

[0189] Example 3 adds a multi-user management module. When the intelligent robot faces interaction requests from multiple users at the same time, it accurately identifies the user's identity through biometric recognition technologies such as face recognition and voice recognition, and sets priority division rules based on the user's identity and the urgency of the task to scientifically determine the priority of each user. Then, the user instructions are arranged into a task queue according to the priority, and the instructions of high-priority users are given priority to ensure that urgent and important tasks can be responded to and processed in a timely manner. After the instruction is executed, the execution result is also fed back to the user to ensure the integrity of the user interaction. This multi-user management mechanism effectively improves the interaction management ability and task processing efficiency of the intelligent robot in multi-user scenarios, avoids the interaction confusion caused by simultaneous requests from multiple users, enables the intelligent robot to meet the needs of different users in a more orderly and efficient manner, and improves the overall experience of multiple users using the intelligent robot.

[0190] As the fourth embodiment of the present invention:

[0191] See also Figure 1 and Figure 2 As shown, when the present application is implemented, compared with Example 1, Example 2 and Example 3, the technical solution of this embodiment is to combine the solutions of the above-mentioned Example 1, Example 2 and Example 3 for implementation.

[0192] Example 4 combines the solutions of Examples 1, 2, and 3, integrating the advantages of each embodiment. It not only has the ability to accurately parse various instructions and effectively process unknown instructions in Example 1, but also has the simple and efficient similarity calculation method in Example 2, and also includes the powerful multi-user management function in Example 3. This has significantly improved the accuracy of instruction processing, computing efficiency, multi-user scenario management, and other aspects of the intelligent robot interaction system. It can meet the user's interaction needs in different scenarios in an all-round and high-quality manner, greatly enhancing the overall performance and practicality of the system, bringing users a more intelligent, efficient, convenient and personalized interaction experience, and making the intelligent robot interaction system more competitive and adaptable in complex practical application environments.

[0193] See also Figure 1 and Figure 2 As shown, the present invention also provides an intelligent robot interaction method based on deep learning, which is implemented by an intelligent robot interaction system based on deep learning. The method includes the following steps:

[0194] Step 1: Interactive instruction analysis:

[0195] Parse and classify the instructions input by the user to determine the instruction type of the user input instruction;

[0196] Step 2: Single instruction processing:

[0197] The interactive instruction parsing module analyzes and classifies the user input instructions, processes the instructions, and determines the execution plan of the intelligent robot;

[0198] Step 3: Multi-command management:

[0199] When the intelligent robot faces interaction requests from multiple users at the same time, it conducts multi-user interaction management analysis to respond to the interaction requests of multiple users one by one.

[0200] It should be stated that all user data collected in this application is collected with the user's consent and authorization, and the use of user data is legal and compliant, and the use and processing of user data complies with the relevant laws, regulations and standards of the relevant regions.

[0201] The above formulas are all dimensionless and numerical calculations. The formulas are obtained by collecting a large amount of data and performing software simulation to obtain the most recent real situation. The preset parameters and thresholds in the formulas are set by technicians in this field according to actual conditions.

[0202] The above description is merely a specific embodiment of the present application, but the scope of protection of the present application is not limited thereto. Any changes or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in this application should be included in the scope of protection of this application. Therefore, the scope of protection of this application should be based on the scope of protection of the claims.

Claims

1. An intelligent robot interaction system based on deep learning, characterized in that: include: The interactive instruction parsing module is used to parse and classify the instructions input by the user by traversing each instruction in a preset instruction database, calculating the similarity between the user input instruction and each instruction in the instruction database, and comparing the obtained similarity with a preset similarity threshold to determine whether the user input instruction is a known instruction or an unknown instruction; The instruction processing module is used to process the user input instructions according to the analysis and classification results of the interactive instruction parsing module, and determine the execution plan of the intelligent robot; The multi-user management module is used to perform multi-user interaction management and analysis when the intelligent robot faces interaction requests from multiple users at the same time, and to respond to the interaction requests of multiple users one by one.

2. The deep learning-based intelligent robot interaction system according to claim 1, characterized in that: The similarity calculation method uses the edit distance algorithm in the text similarity calculation method, as follows: Extract the character length M0 of the user input instruction D0 and each instruction D in the instruction database i The character length M i , i = 1, 2, ... n, n is the total number of instructions in the instruction database; Construct a (M0+1)×(M i +1) of the matrix E=[E(uv i )]; where E(uv i ) means the first u characters of D0 and D i The front v i The edit distance between characters, and 1≤u≤M0, 1≤v i ≤M i ; When E(00)=0, it means the edit distance between two empty strings is 0; For u = 1 to M0, E(u0) = u, which means that it takes u deletion operations to transform the first u characters of D0 into an empty string; For v i =1 to M i ,E(0v i )=v i , which means transforming the empty string into D i The front v i characters need v i Insert operation; For u=1 to M0, v i =1 to M i , if the uth character of D0 is the same as D i No. v i characters are the same, then: E(uv i )=min{E(u-1,v i -1),E(u-1,v i )+1,E(u,v i -1)+1}; Among them, E(u-1,j-1) means that the current characters are the same and no additional operation is required. E(u-1,v i )+1 means deleting the uth character of D0, E(u,v i -1)+1 means insert D into I i No. v i characters; If the uth character of D0 is the same as D i No. v i characters are different, then: E(uv i )=min{E(u-1,v i -1)+1,E(u-1,v i )+1,E(u,v i -1)+1}; Among them, E(u-1,v i -1)+1 means replace the uth character of D0 with D i No. v i characters; pass Calculate the difference between the user input command D0 and the command D in the command database i The similarity S(D0, D i ); When S(D0, D i )≥T, then the instruction D0 is determined to be a known instruction; otherwise, the instruction D0 is determined to be an unknown instruction.

3. The deep learning-based intelligent robot interaction system according to claim 1, characterized in that: The similarity calculation method also uses a simple similarity calculation method based on character matching, as follows: First, extract the character length M0 of the user input instruction D0 and each instruction D in the instruction database i The character length M i ; Then define a counter C=0 to record the number of matching characters; From D0 and D i Comparison starts with the first character: When comparing to the e-th character, if the e-th character of D0 is the same as D i The e-th character is the same as that of , then C=C+1; Where, e=1, ...min(M0,M i ); After completing the comparison, pass: Calculate the difference between the user input command D0 and the command D in the command database i The similarity S(D0, D i ).

4. The deep learning-based intelligent robot interaction system according to claim 1, characterized in that: The instructions are processed as follows: When it is determined that instruction D0 is a known instruction, the instruction input by the user is executed according to the execution scheme of the known instruction with the greatest similarity to instruction D0; When it is determined that the instruction D0 is an unknown instruction, a depth matching process is performed on the instruction D0, and then an execution plan of the user input instruction is determined based on the depth matching process.

5. The deep learning-based intelligent robot interaction system according to claim 4, characterized in that: Depth matching is handled as follows: The keyword extraction method based on word frequency statistics is used to obtain the keyword set K of D0 in the unknown instruction = {K z }, where z = 1, 2, ... t, t is the number of keywords; In the pre-set extended solution database, the solution related to the keyword set K is retrieved, and each solution in the extended solution database is recorded as F j , j = 1, 2, ... m, m is the total number of records of treatment solutions in the extended treatment solution database DB; Then, D0 in the unknown instruction and each processing solution F j The matching degree is calculated as follows: Extract each treatment plan F j The keyword set K contained in j ; Calculate K and K first j The number of intersection elements Na, that is, Na=|K∩K j ∣; Then calculate K and K j The number of elements in the union of Nb, that is, Nb = |K∪K j ∣; Then through: Calculate D0 and each processing solution F in the unknown instruction j The matching degree M(D0,F j ); Then D0 in the unknown instruction and each processing solution F j The matching degree M(D0,F j ) is compared with the pre-set matching threshold My: If there exists a condition that satisfies M(D0,F j )≥My, then select the corresponding solution F j to handle the unknown instruction.

6. The deep learning-based intelligent robot interaction system according to claim 5, characterized in that: If there is no such condition that M(D0,F j )≥My, the robot will execute the following strategies, as follows: The keyword set K is recombined and expanded, and a new instruction description set C is generated by arranging and combining the keywords in different orders. g }, where g = 1, 2, ..., s, where s is the number of instruction descriptions in set C; For each instruction description C in the instruction description set C g , perform an association search in the pre-established general knowledge graph. In the association search process, first parse C g Keywords and key concepts in the knowledge graph; based on the pre-built relationship paths in the knowledge graph, starting from these keywords, traverse along various relationship edges; the robot gradually expands the search scope along the relationship edges to find g Related processing suggestions or information, forming a related information set R g ; From R g The most relevant information is selected and the relevance score is based on the number of keyword matches. Count the number of keywords that match the keyword set K in each information node, and use it to calculate the number of keywords in R g Sort all the information nodes in the , and then give priority to the information nodes with high ranking and high score, and get a temporary solution for the unknown instruction; Show the user a temporary solution and ask "Do you want to implement this solution?" If the user confirms, the temporary solution will be implemented; If the user denies the request, the system will request the user to provide more instruction details or related information, and then re-parse, classify and process the instructions input by the user.

7. The deep learning-based intelligent robot interaction system according to claim 1, characterized in that: The multi-user interaction management and analysis methods are as follows: Step G1: First, user identification is performed. Specifically, the identity information of each user is determined by biometric identification means corresponding to face recognition technology and voice recognition technology, and the user identity information set is formed as U = {U f }, f = 1, 2, ... h, h is the number of users; Step G2: After identifying the user identity, determine the user priority according to the pre-set priority classification rules; Step G3: Arrange the commands input by each user into a task queue according to the user priority from high to low. In the task queue, the interactive instruction parsing module first parses the highest priority user input instruction in the task queue to determine whether it is a known instruction or an unknown instruction; If the command is known, the command input by the user is directly executed according to the corresponding execution plan; If it is an unknown instruction, the instruction processing module performs unknown instruction processing on the instruction processing method for the unknown instruction; After the instruction is executed, the completed instruction is removed from the task queue and the next highest priority task is processed until the task queue is empty.

8. The deep learning-based intelligent robot interaction system according to claim 7, characterized in that: in, The prioritization rules include: Different users are assigned different basic priority values A1 according to their identities; Where A1={A1 x1 }, x1 = 1, 2, ..., q1, q1 refers to the number of attributes corresponding to the user identity; Different users are assigned different emergency priority values A2 according to the urgency of the task corresponding to the user input instruction; Where A2={A2 x2 }, x2=1, 2, ... q2, q2 refers to the number of task types corresponding to the urgency of the task.

9. The deep learning-based intelligent robot interaction system according to claim 8, characterized in that: User priority is determined as follows: Extract the basic priority value A1 corresponding to each user f and emergency priority value A2 f ; Then passed: P f =A1 f ×α1+A2 f ×α2, calculate the user priority P f ; Where α1 and α2 are preset weight coefficients corresponding to the basic priority value and the emergency priority value.

10. A deep learning-based intelligent robot interaction method, the method being implemented by a deep learning-based intelligent robot interaction system according to any one of claims 1 to 9, characterized in that: The method comprises the following steps: Interactive instruction parsing: parse and classify the instructions input by the user to determine the instruction type of the user input instruction; Single instruction processing: The interactive instruction parsing module analyzes and classifies the user input instructions, processes the instructions, and determines the execution plan of the intelligent robot; Multi-command management: When the intelligent robot faces interaction requests from multiple users at the same time, it performs multi-user interaction management analysis to respond to the interaction requests of multiple users one by one.

Images