JP2019040605A - Feeling interactive method based on humor creation and robot system - Google Patents

Abstract

To disclose a feeling interactive method based on humor creation and a robot system.SOLUTION: A user's data is acquired, a humor paragraph is created by the user's data and outputted, a user's feeling feedback to the outputted humor paragraph is monitored, when the user's feeling feedback is identified as a default feeling, the outputted humor paragraph is determined to have a humor characteristic, the larger the default feeling level is as a result of identifying the user's feeling feedback, the larger the humor level of the outputted humor paragraph is, when the user's feeling feedback is identified to be a feeling other than the default feeling, the outputted humor paragraph is determined to have no humor characteristic, and the humor paragraph having the humor characteristic, environment information corresponding to the humor paragraph and the humor level are stored in a humor big data knowledge base.SELECTED DRAWING: Figure 1

Description

The present invention belongs to the emotional interactive method and system, particularly the emotion interactive method and system based on humor generation, and the artificial intelligence technical field.

With the rapid development of robot technology, user demands for robot functions increase, and emotion and humor are one of the most important contents of wisdom robots.

The main function of the existing emotion calculation is to identify and express emotions, and the main function of the existing humor calculation is to recognize and generate humor, but the existing emotion calculation technology and humor calculation technology are combined. Without progressing to each.

Existing robots also have the following problems:

1) Since an existing robot acquires a humor paragraph or humor component through a humor knowledge base and the number of humor paragraphs and components in the humor knowledge base is limited, the level and level of humor are low.

2) Although existing robots can sense, identify, understand, and express emotions, but cannot predict human emotions, they do not know how human emotions can be changed and cannot please users. The robot cannot predict what emotions will be generated by the user because of different responses to the same paragraph of the user, for example, what responses will default the user's emotions, what responses will offend the user, and therefore the robot Can't choose a reply to create a particular emotion for the user.

Chinese Patent Application No. 103488293

  The object of the present invention is to provide an emotion interaction method based on humor generation for solving the problems submitted to the background art described above, which method can be inspired and changed by user's emotion generation through humor and through emotion identification Test the humor effect, and then answer further to generate specific emotions for the user.

Another object of the present invention is to provide an emotion dialogue method and robot system based on humor generation.

The objectives of the present invention can be achieved with the following technical plan:

An emotional dialogue method based on humor generation, which method is applied to a robot and includes:
Get user data,
Generate and output humor paragraphs based on user data,
Monitor the user's emotional feedback to the output humor paragraph,
When the user's emotion feedback is identified as the default emotion, the output humor paragraph is judged to have humor characteristics, and the higher the default emotion level in the user's emotion feedback identification result, the greater the output humor paragraph. The level of humor increases,
If the user ’s emotional feedback is identified as an emotion other than the default emotion, the output humor paragraph will refrain from having humor characteristics,
A humor paragraph having humor characteristics, environmental information corresponding to the humor paragraph, a humor level of the humor paragraph, and a user's emotion feedback for the humor paragraph are stored in a humor big data knowledge base.

Further, generating and outputting a humor paragraph based on the user's data specifically includes:
Extract some data from the user ’s data as the first small data, and some other data as the second small data,
Get big data,
Obtain the data related to the first small data from the big data and clean it to make the first correlation data set,
Mining a paragraph in which the positive correlation with the first small data from the first correlation data set is greater than the default positive correlation threshold, storing the mined paragraph and its positive correlation in a positive correlation paragraph set,
Mining a paragraph whose negative correlation with the second small data is greater than the default negative correlation threshold from the positive correlation paragraph set and storing the mined paragraph and its positive and negative correlation degrees in the positive and negative correlation paragraph set ,
Calculate the humor level of the corresponding paragraph based on the positive correlation and negative correlation corresponding to each paragraph of the positive / negative correlation paragraph set, and save the paragraph whose humor level is higher than the default humor threshold in the humor paragraph set,
The humor paragraph with the highest humor level is selected from the humor paragraph set, and the corresponding humor paragraph is output.

Further, generating and outputting a humor paragraph based on the user's data specifically includes:
Extract some data from the user ’s data as the first small data, and some other data as the second small data,
Get big data,
Obtain the data related to the first small data from the big data and clean it to make the first correlation data set,
Mining a paragraph in which the negative correlation with the second small data from the first correlation data set is greater than the default negative correlation threshold, storing the mined paragraph and its negative correlation in a negative correlation paragraph set,
Mines the negative correlation paragraph set that has a positive correlation with the first small data that is greater than the default negative correlation threshold, and saves the mined paragraph and its negative and positive correlation degrees in the negative positive correlation paragraph set. And
Calculate the humor level of the corresponding paragraph based on the negative correlation and the positive correlation corresponding to each paragraph of the negative positive correlation paragraph set, and save the paragraph whose humor level is higher than the default humor threshold in the humor paragraph set,
The humor paragraph with the highest humor level is selected from the humor paragraph set, and the corresponding humor paragraph is output.

Further, the method includes:
Based on the data in the humor big data knowledge base and the emotion feedback other than the default emotion of different types of users to different responses each time in the past, the emotion feedback related information of different types of users to different responses in the past is obtained and The emotion feedback related information includes user information, reply content, emotion feedback content and emotion feedback type,
Store emotional feedback related information of different types of users for different responses in the past in emotion feedback big data,
Get user information waiting for a reply, get a candidate reply set for users waiting for a reply,
Get anticipatory user sentiment feedback,
Find all sentiment feedback records that most closely match the sentiment feedback big data to the user information awaiting reply and the reply content in the candidate reply set to the waiting reply user,
Extract emotional feedback from emotional feedback records, calculate the degree of agreement between emotional feedback and expected user emotional feedback,
Based on the emotion feedback record corresponding to the greatest degree of agreement between the emotion feedback and the expected user emotion feedback, the corresponding response content of the emotion feedback record is obtained and the corresponding response content is returned to the user waiting for the response .

In addition, searching for all emotion feedback records that most closely match the user information and the reply content in the candidate reply set from the emotion feedback big data, specifically:
User information for each emotional feedback record in emotional feedback big data, match the response content to the user information waiting for the response, the response content in the candidate answer set for the user waiting for the response, each emotion feedback record and the user information waiting for the response , Get the degree of match with the candidate reply set for the user waiting for reply, select the corresponding emotion feedback record with the largest match from among them, select the user information waiting for reply and the candidate reply to the user waiting for reply The emotional feedback record that best matches the set.

Another object of the invention can be achieved with the following technical plan:

Emotion interaction method and robot system based on humor generation, said system comprising user data acquisition module, humor paragraph generation output module, emotion feedback monitoring module, first determination module, second determination module and humor big data knowledge base acquisition module Including
The user data acquisition module acquires user data,
The humor paragraph generation / output module generates and outputs a humor paragraph according to user data,
The emotion feedback monitoring module monitors the user's emotion feedback to the output humor paragraph,
When the user's emotional feedback is identified as the default emotion, the first determination module determines that the output humor paragraph has humor characteristics, and the user's emotional feedback identification results in a high default emotion level. The higher the humor level of the output humor paragraph,
The second determination module determines that the output humor paragraph has no humor characteristic when the user's emotion feedback is identified as an emotion other than the default emotion,
The humor big data knowledge base acquisition module includes humor paragraphs having humor characteristics, environmental information corresponding to the humor paragraphs, humor levels of the humor paragraphs, and user emotion feedback to the humor paragraphs. Save to.

Further, the humor paragraph generation output module includes a small data extraction unit, a big data acquisition unit, a first correlation data set acquisition unit, a positive correlation paragraph set acquisition unit, a positive / negative correlation paragraph set acquisition unit, a humor paragraph set acquisition unit, and a humor paragraph output Including units,
The small data extraction unit extracts some data from the user's data as first small data, and extracts some data as second small data,
The big data acquisition unit acquires big data,
The first correlation data set acquisition unit obtains data related to the first small data from the big data and cleans it to obtain a first correlation data set,
The positive correlation paragraph set acquisition unit mines the paragraph in which the positive correlation with the first small data from the first correlation data set is greater than the default positive correlation threshold, and corrects the mined paragraph and its positive correlation. Save in a set of correlated paragraphs,
The positive / negative correlation paragraph set acquisition unit mines a paragraph whose negative correlation with the second small data is larger than the default negative correlation threshold from the positive correlation paragraph set, and the mined paragraph and its positive correlation and negative correlation Store degrees in a positive and negative correlation paragraph set,
The humor paragraph set acquisition unit calculates the humor level of the corresponding paragraph based on the positive correlation and the negative correlation corresponding to each paragraph of the positive / negative correlation paragraph set, and the humor level is greater than the default humor threshold. Save it in a collective,
The humor paragraph output unit selects a humor paragraph having the highest humor level from the set of humor paragraphs, and outputs the corresponding humor paragraph.

The humor paragraph generation / output module includes a small data extraction unit, a big data acquisition unit, a first correlation data set acquisition unit, a negative correlation paragraph set acquisition unit, a negative positive correlation paragraph set acquisition unit, a humor paragraph set acquisition unit, and a humor Including a paragraph output unit,
The small data extraction unit extracts some data from the user's data as first small data, and extracts some data as second small data,
The big data acquisition unit acquires big data,
The first correlation data set acquisition unit obtains data related to the first small data from the big data and cleans it to obtain a first correlation data set,
The negative correlation paragraph set acquisition unit mines a paragraph in which the negative correlation with the second small data from the first correlation data set is greater than the default negative correlation threshold, and negatives the mined paragraph and its negative correlation. Save in a set of correlated paragraphs,
The negative positive correlation paragraph set acquisition unit mines a paragraph in which the positive correlation degree with the first small data is larger than the default positive correlation degree threshold from the negative correlation paragraph set, the mined paragraph and its negative correlation degree and the positive correlation degree. Save the degree of correlation in the negative positive correlation paragraph set,
The humor paragraph set acquisition unit calculates a humor level of a corresponding paragraph based on a negative correlation level and a positive correlation level corresponding to each paragraph of the negative positive correlation paragraph set, and calculates a humor level greater than a default humor threshold. Save it in a humor paragraph set,
The humor paragraph output unit selects a humor paragraph having the highest humor level from the set of humor paragraphs, and outputs the corresponding humor paragraph.

Further, the system includes an emotion feedback related information acquisition module, an emotion feedback big data acquisition module, a candidate answer set acquisition module, an expected user emotion feedback acquisition module, an emotion feedback record search module, a matching score calculation module, and an answer content acquisition module ,
The sentiment feedback related information acquisition module is based on data in the humor big data knowledge base and sentiment feedback other than the default sentiment of different types of users to different responses each time in the past, based on different types of users' responses to different responses in the past. Emotion feedback related information is acquired, and the emotion feedback related information includes user information, reply content, emotion feedback content and emotion feedback type.
The emotion feedback big data acquisition module stores emotion feedback related information of different types of users to different responses each time in the past in emotion feedback big data,
The candidate answer set acquisition module acquires user information waiting for a reply, acquires a candidate answer set for a user waiting for a reply,
The anticipation user emotion feedback acquisition module acquires anticipation user emotion feedback,
The sentiment feedback record search module searches all sentiment feedback records that most closely match the reply information in the candidate answer set to the waiting user information and the waiting reply user information from the sentiment feedback big data,
The coincidence calculation module extracts emotion feedback from the emotion feedback record, calculates the agreement between the emotion feedback and the expected user emotion feedback,
The reply content acquisition module acquires the reply content corresponding to the corresponding emotion feedback record based on the emotion feedback record corresponding to the greatest degree of coincidence between the emotion feedback and the expected user emotion feedback, and waits for the corresponding reply content Answer to the user.

Further, the emotion feedback record retrieval module is specifically:
User information for each emotional feedback record in emotional feedback big data, match the response content to the user information waiting for the response, the response content in the candidate answer set for the user waiting for the response, each emotion feedback record and the user information waiting for the response , Get the degree of match with the candidate reply set for the user waiting for reply, select the corresponding emotion feedback record with the largest match from among them, select the user information waiting for reply and the candidate reply to the user waiting for reply The emotional feedback record that best matches the set.

The present invention has the following beneficial effects on existing technologies:

1. The present invention inspired and changed the user's emotion through the output of the humor paragraph, and monitored the user's emotion feedback to the output humor paragraph, that is, verified the effect of humor by emotion identification and fed back to the user The identified humor paragraph is a humor paragraph corresponding to the default emotion, so it is determined that it has humor characteristics, and the humor paragraph having humor characteristics and related information are stored in the humor big data knowledge base. Then select a humor paragraph through the future humor big data knowledge base, thereby improving the robot's ability to generate humor.

2. The present invention first performs a positive correlation analysis on the big data and the first small data, obtains a positive correlation paragraph set, mines a paragraph negatively correlated with the second small data from the positive correlation paragraph set, Save the mined paragraphs and their corresponding positive and negative correlation degrees in the positive and negative correlation paragraph set, get the humor paragraph set by the positive and negative correlation paragraph set, and select the humor paragraph with the highest humor level from the humor paragraph set However, the humor level of the robot is improved, and the positive and negative correlation paragraphs are the same, so the relevant paragraph is rational and funny, and has a humorous effect. Sort the output humor paragraphs, humor big data It provides a good choice by identification based, to further improve the humor generating capability of the robot.

3. The present invention first performs a negative correlation analysis on the big data and the second small data, obtains a negative correlation paragraph set, mines a paragraph positively correlated with the second small data from the negative correlation paragraph set, Save the mined paragraphs and their corresponding negative and positive correlations in the negative positive correlation paragraph set, obtain the humor paragraph set from the negative positive correlation paragraph set, and the humor paragraph with the highest humor level from the humor paragraph set To improve the robot's humor level, and the negative and positive correlation paragraphs are the same, so that the paragraph is reasonable and funny, and has a humorous effect, and then the user emotion identification in the subsequent step To sort the output humor paragraphs, humor big data It provides a good choice by identification based, to further improve the humor generating capability of the robot.

4. The present invention obtains emotion feedback big data based on data in the humor big data knowledge base and emotion feedback other than the user's default emotion in the past to different replies. Can search for all emotion feedback records that best match the response of a candidate response to a user waiting for a response, and predict what emotions the user will have if the user responds differently to the same word In addition, it is possible to make the user reply with a certain feeling according to the prediction result.

The present invention will be further described with reference to the accompanying drawings. However, the embodiment of the accompanying drawings is not limited to the present invention, and for the ordinary technical field of this field, other than the creative labor, Acquire the attached figure.

FIG. 1 is a flowchart of an emotion dialogue method based on humor generation according to Embodiment 1 of the present invention. FIG. 2 is a flowchart for generating and outputting a humor paragraph according to the second embodiment of the present invention. FIG. 3 is a flowchart for mining the positive correlation paragraph according to the second embodiment of the present invention. FIG. 4 is a flowchart for mining the negative correlation paragraph according to the second embodiment of the present invention. FIG. 5 is a simplified sketch of acquisition of a humor paragraph set according to the second embodiment of the present invention. FIG. 6 is a simplified sketch for generating a humor paragraph based on the positive / negative correlation analysis of big data according to the second embodiment of the present invention. FIG. 7 is a flowchart for generating and outputting a humor paragraph according to the third embodiment of the present invention. FIG. 8 is a flowchart for mining the negative correlation paragraph according to the third embodiment of the present invention. FIG. 9 is a flowchart for mining the positive correlation paragraph according to the third embodiment of the present invention. FIG. 10 is a simple sketch for acquiring a humor paragraph set according to the third embodiment of the present invention. FIG. 11 is a simplified sketch for generating a humor paragraph based on the negative / positive correlation analysis of big data according to the third embodiment of the present invention. FIG. 12 is a flowchart of an emotion dialogue method based on humor generation according to Embodiment 4 of the present invention. FIG. 13 is a block diagram of an emotion dialogue robot system based on humor generation according to Embodiment 5 of the present invention. FIG. 14 is a block diagram of a humor paragraph generation / output module according to Embodiment 6 of the present invention. FIG. 15 is a block diagram of a humor paragraph generation / output module according to the seventh embodiment of the present invention. FIG. 16 is a block diagram of an emotion dialogue robot system based on humor generation according to Embodiment 8 of the present invention.

In the following, the embodiment of the present invention will be described in more detail with reference to the accompanying drawings, but the embodiment of the present invention is not limited to this.

Example 1:

The robot's output humor paragraph is generated by the humor calculation, and it is identified by the emotion calculation that the user who reads or listens to the corresponding humor paragraph laughs out. At this time, the user who reads or listens to the corresponding humor paragraph Having identified that he laughed, he found that the humor paragraph output by the robot passed, and this successful experience will be applied to future robot humor generation.

As FIG. 1 shows, this embodiment provides an emotion dialogue method based on humor generation, which can be applied to a robot, and includes the following steps:

S101, obtaining user data.

In this embodiment, the robot refers to an intelligent robot, which is a machine that can simulate human actions and thoughts, and includes chat programs and other computer programs. For example, a camera is attached to the robot's eye, the user's video image data can be obtained, a pucker is attached to the robot's ear, the user's audio data can be obtained, and other data The sensor acquires sense data (for example, user's body temperature, heart rate, etc.), and each terminal acquires text data (for example, user attributes, paragraphs, etc.) input by the user.

S102, generating and outputting a humor paragraph based on user data.

In the present embodiment, a paragraph input by a user is acquired, a humor paragraph is generated and output through an existing humor generation method, and a method of outputting is text or voice.

S103, monitoring the emotion feedback of the user to the outputted humor paragraph.
The user's emotional feedback to the output humor paragraph is monitored, and the robot uses the camera to inspect and measure the facial expression of the user, to inspect and measure the user's voice with a microphone, For example, when monitoring whether the user's emotion feedback to the output humor paragraph is the default emotion, the camera can be used to test whether the user is smiling, Engineers in this field can measure whether or not there is a laughter in the user's voice using text, or test and measure whether the text entered by the user contains humor or haha using text input. Can understand that there is a judgment method corresponding to other emotions besides the default emotion.

S104, when the emotion fed back to the user is identified as the default emotion, the output humor paragraph is determined to have humor characteristics, and the level of the default emotion is included in the identification result of the emotion fed back to the user. The higher the value, the higher the humor level of the output humor paragraph, and if the emotion fed back to the user is identified as an emotion other than the default emotion, the output humor paragraph does not have humor characteristics. The

The default emotion setting method is as follows: The emotion fed back to the actual humor paragraph is inspected and measured as the default emotion. For example, you can test and measure the emotions that 10,000 people have fed back against 100 actual humor paragraphs, make the most emotional feedback the default emotion, and of course you can get the default emotion in other ways, Further, the default emotion may be a prorated combination of various emotions. The so-called actual humor paragraph refers to a paragraph with a high humor level. If the default emotion is not set, the emotion “fun” is automatically set as the default emotion, because when a human hears a humor paragraph, the most common reaction is to laugh, and laughing is a fun emotion It is a kind.

In this embodiment, the default emotion is set to “fun”, and the level of the default emotion is a level of fun, which is determined by any one of the width of smile, the decibel of laughter and the number of characters related to “laugh” in the text. Or determined by a combination of two or three of them, and depending on the level of fun, the corresponding humor level is high, medium, low or first grade, second grade, third grade, fourth grade, etc. or A grade, B grade , C class, D class, etc., and humor level can be divided by scoring. Emotions other than default emotions include silence, anger, and pain.

S105, based on the paragraph input by the user anew, the non-humor paragraph is output by the normal reply phrase generation method, and the user is told, “The previous is a humor paragraph, please do not mind, please refer to this non-humor paragraph now In reference".

The step can be omitted during the experiment, but it is very important in practical applications, and lack of the step can lead to user misunderstanding, which is generally a humor paragraph mistake that makes people laugh Through the corresponding steps, the user can obtain a pleasant experience and correct knowledge by judging whether the correctness is correct. In particular, when applying to an educational support robot, the corresponding step is further indispensable.

S106, the humor paragraph having the characteristics of humor, the environmental information corresponding to the humor paragraph, the humor level of the humor paragraph, and the user's emotion feedback for the humor paragraph are stored in the humor big data knowledge base.
The environmental information of the corresponding step refers to the time, place, and both attributes when the user interacts with the robot. When the robot generates humor in the future, each humor paragraph, each humor paragraph in the humor big data knowledge base The humor paragraph is selected based on the corresponding environmental information, the humor level of each humor paragraph, and the user's emotional feedback for each humor paragraph, and the robot's ability to generate humor is improved.

Example 2:

The other steps of the present embodiment are the same as those of the first embodiment, and different portions are realized by the method in which step S102 is based on the positive / negative correlation analysis of big data. Specifically, as shown in FIG. Including:

S1021, part of data is extracted as first small data from user data, and another part of data is extracted as second small data.

“Throw” some small data (stones) into big data (pond), generate humor (the ripples spread at once with stones), and the process of identifying or generating humor is between small data and big data Is associated data mining, thus creating humor that cannot be identified or generated with only small data or big data, which is the result of robot humor identification and generation based on big data and first small data.

Small data is personalized data unique to the user's dialogue process, and multiple small data can be acquired through different paths during the same dialogue process, such as user paragraphs, user attributes, dialogue scenes, Dialogue time, dialogue place, robot attributes, etc.

Among these small data, the user's paragraph is the core small data, and if the robot's response completely negates the user's paragraph (that is, the response that is completely negatively correlated with the user's paragraph, that is, the dialogue does not fit) Affirming all or some of the user ’s paragraphs (ie, the user ’s paragraph) in order to not only have no humor effect, but also to make the user feel that the robot is speaking Response that is positively correlated to the paragraph), which resonates the user and forms the basis for humor.

In this embodiment, all or part of the user's paragraph is extracted as the first small data from the user data acquired in step S101. These small data are the basis of humor, but it is impossible to affirm all sensed small data (ie, a response that is fully positively correlated to all sensed small data). If it is affirmed, it becomes a normal response paragraph and does not become humor, so it is necessary to negate some small data in it, and it is necessary to integrate this negative part with the affirmative part and fuse it There is a realization of humor of unity.

S1022, obtaining big data.

Big data is data accumulated over a long period of time by a large number of users. For example, Internet big data is data accumulated over a long period of time by a myriad of Internet users. Therefore, big data is external data for robots and dialogue processes. In this embodiment, the robot can acquire big data through the server.

In step S1023, data related to the first small data is obtained from the big data and cleaned to obtain a first correlation data set.
Get data related to the first small data from the big data: Search the paragraph where the degree of coincidence with the first small data is higher than the default coincidence threshold from the big data, and the paragraph with the big data and the small data in it The degree of coincidence indicates that the number of small data words shown in the corresponding paragraph is divided by the number of small data words.

Clean data: The data may be a single paragraph consisting of multiple words, marked with a word that appears in a small data keyword, and marked with one word and the previous or next word. If not, delete the word or mark the word in the small data keyword,
If a word is not marked, delete the word,
Among them, a phrase is a phrase that ends with a punctuation mark, exclamation mark, question mark, or other end-of-sentence identifier.

S1024: Mining a paragraph whose positive correlation degree with the first small data from the first correlation data set is larger than the default positive correlation threshold, and storing the mined paragraph and its positive correlation degree in a positive correlation paragraph set.

Paragraphs whose positive correlation is greater than the default positive correlation threshold are called positive correlation paragraphs, and the mining of positive correlation paragraphs can also be realized with the technique of correlation analysis of existing big data, and can also be realized with the method of FIG. Specifically, it includes the following steps:

S10241, obtain the number of similar paragraphs of the first small data in the big data as the first quantity, record in m,

S10242, the number of similar data of each data in the first correlation data set in the big data is statistically recorded as the second quantity of the corresponding data, and recorded by n,

S10243, the number of similar data in the big data of the data acquired after merging the corresponding data and the first small data is statistically recorded as the third quantity of the corresponding data, recorded in o,

  S10244, based on the first quantity m, the second quantity n and the third quantity o of the relevant data, calculate the degree of positive correlation of the relevant data and record it in pc,

Specifically, the calculation formula for the positive correlation degree pc of the corresponding data is as follows:

  pc = o / ((m × n) 1/2) (1)

S10245, determining whether the positive correlation between all the paragraphs of the first correlation data set and the second small data is greater than the default positive correlation threshold;

S10246, the paragraph whose positive correlation degree with the first small data is larger than the default positive correlation degree threshold and the positive correlation degree are stored in the positive correlation paragraph set and recorded in Z1.

If the positive correlation is sufficiently large (ie, greater than the default positive correlation threshold), the combination of the relevant paragraph and the first small data is the consensus of most people, thereby making the humor paragraph rational.

S1025, mining a paragraph whose negative correlation with the second small data is larger than the default negative correlation threshold from the positive correlation paragraph set, and converting the mined paragraph and its positive correlation degree and negative correlation degree into a positive / negative correlation paragraph set. save.

Paragraphs whose negative correlation is greater than the default negative correlation threshold are called negative correlation paragraphs, and mining of negative correlation paragraphs can be realized with the technology of analyzing the correlation of existing big data, and can also be realized with the method of FIG. Specifically, it includes the following steps:

S10251, the second small data obtains the number of similar paragraphs in the big data, and records it as p as the fourth quantity,

S10252, each piece of data in the positive correlation paragraph set Z1 statistically counts the number of similar data in the big data, sets the fifth quantity of the corresponding data, records with q,

S10253, the data obtained by merging the corresponding data and the second small data is the number of similar data in the big data, and recorded as r as the sixth quantity of the corresponding data,

S10254, based on the fourth quantity p, the fifth quantity q and the sixth quantity r of the relevant data, calculate the negative correlation of the relevant data, and record it as nc,

Specifically, the calculation formula for the negative correlation nc of the corresponding data is as follows:

nc = 1-r / ((p × q) 1/2) (2)

S10255, determining whether the negative correlation with the second small data of all the paragraphs in the positive correlation paragraph set Z1 is greater than the default negative correlation threshold;

S10256, the paragraph in which the negative correlation with the second small data is greater than the default negative correlation threshold and the corresponding positive correlation and negative correlation are stored in the positive / negative correlation paragraph set and recorded in Z2.

If the negative correlation is large enough (ie, greater than the default negative correlation threshold), the combination of the relevant paragraph and the second small data is the non-common sense of most people, thereby making the humor paragraph funny.

In step S1026, the humor level of the corresponding paragraph is calculated based on the positive correlation and the negative correlation corresponding to each paragraph of the positive / negative correlation paragraph set, and the paragraph having a humor level larger than the default humor threshold is stored in the humor paragraph set.

In this embodiment, one paragraph and its positive correlation and negative correlation are obtained from the positive / negative correlation paragraph set Z2, and the positive correlation is 80% and the negative correlation is 90%. The humor level of the corresponding paragraph is calculated based on the correlation degree and the negative correlation degree, and the paragraph in which the humor level is higher than the default humor threshold and the humor level are stored in the humor paragraph set and recorded as “Z”.

Specifically, the formula for calculating the humor level h in the relevant paragraph is as follows:

h = (pc × nc) 1/2 (3)

The quantity statistics in steps S1024 and S1025 are to adopt an existing mapreduce model and perform quick statistics for big data using hadoop or spark.

FIG. 5 is a simple sketch for obtaining a humor paragraph set. The big data analysis process to be adopted by mining a paragraph set Z1 that coincides with the first small data y1 from the big data X (positive correlation) is Z1 = f (X, Record at y1), which corresponds to a positive correlation step. The big data analysis process adopted by mining the paragraph set Z that matches the first small data y1 from the big data X (positive correlation) and does not match the second small data y2 (negative correlation) is g (X, y2 ∪f (X, y1)), which corresponds to the conflicting unified staircase of conflicting unified humor, that is, the stair that climbs up to the humor identification and generation staircase. The process of these two steps is not divided, but influences each other and is organized organically.

FIG. 6 is a simplified sketch of the above steps S1021 to S1026. A positive correlation analysis is performed on the big data X and the first small data y1, Z1 is mined from the big data, and then the big data X of Z1 and the second data are obtained. A negative correlation analysis is performed on the small data y2, and Z is mined and acquired from the big data.

S1027, the humor paragraph with the highest humor level is selected from the humor paragraph set, and the corresponding humor paragraph is output.

In this example, a humorous paragraph is acquired based on the positive / negative correlation analysis of big data, the humor level of the robot is improved, and since the positive correlation paragraph and the negative correlation paragraph are the same, the corresponding paragraph is rational. It is funny and forms a humor effect, and through user emotion identification in the subsequent steps of Example 1, the humor paragraph generated based on the positive and negative correlation analysis of big data is screened, and a better choice for the humor big data knowledge base Provide and further improve the humor generation ability of the robot.

Example 3:

The other steps of the present embodiment are the same as those of the first embodiment, and different portions are realized by the method in which step S102 is based on the negative / positive correlation analysis of big data. Specifically, as shown in FIG. including:

S1021, part of the data is extracted as the first small data from the user data, and the other part of the data is extracted as the second small data. The specific contents of this step can be referred to the second embodiment. .

S1022, big data is acquired, and the specific contents of this step can be referred to the above-described second embodiment.

In step S1023, data related to the first small data is obtained from the big data and cleaned to obtain the first correlation data set. The specific contents of this step can be referred to the above-described second embodiment.

S1024: Mining a paragraph in which the negative correlation with the second small data from the first correlation data set is greater than the default negative correlation threshold, and storing the mined paragraph and its negative correlation in a negative correlation paragraph set.

Paragraphs whose negative correlation is greater than the default negative correlation threshold are called negative correlation paragraphs. Mining of negative correlation paragraphs can be realized with the technology of analyzing the correlation of existing big data, and can also be realized through the method of FIG. Specifically, it includes the following steps:

  S10241, the second small data obtains the number of similar paragraphs in big data and records it as m as the first quantity,

S10242, statistics the number of similar data of each data in the first correlation data set in the big data, recorded as n as the second quantity of the corresponding data,

S10243, the data acquired after merging the corresponding data with the second data is the number of similar data in the big data, the third quantity of the corresponding data, recorded in o,

S10244, based on the first quantity m, the second quantity n and the third quantity o of the relevant data, calculate the negative correlation of the relevant data, record it as nc,

Specifically, the calculation formula for the negative correlation nc of the corresponding data is as follows:
nc = 1−o / ((m × n) 1/2) (4)

S10245, determine whether the negative correlation of all paragraphs of the first correlation data set and the second small data is greater than the default negative correlation threshold;

S10246, the paragraph whose negative correlation degree with the second small data is larger than the default threshold and the negative correlation degree are stored in the negative correlation paragraph set, and recorded in Z1.

S1025, mining a paragraph whose positive correlation with the first small data is larger than the default negative correlation threshold from the negative correlation paragraph set, and mining the paragraph and its negative correlation degree and positive correlation degree as a negative positive correlation paragraph set. Save to.

Paragraphs whose positive correlation is greater than the default positive correlation threshold are called positive correlation paragraphs, and mining of positive correlation paragraphs can be realized with the technology of analyzing the correlation of existing big data, and can also be realized through the method shown in FIG. Specifically, it includes the following steps:

S10251, obtain the number of similar paragraphs of the first small data in the big data, and record it as p as the fourth quantity,

S10252, each data of the negative correlation paragraph set Z1 statistically counts the number of similar data in the big data, sets the fifth quantity of the corresponding data, records with q,

S10253, the data obtained by merging the corresponding data with the first small data is the number of similar data in the big data, and recorded as r as the sixth quantity of the corresponding data,

S10254, based on the fourth quantity p, the fifth quantity q and the sixth quantity r of the corresponding data, calculate the degree of positive correlation of the corresponding data, and record it as pc.

Specifically, the calculation formula for the positive correlation degree pc of the corresponding data is as follows:

pc = r / ((p × q) 1/2) (5)

S10255, determining whether the positive correlation between all the paragraphs in the negative correlation paragraph set Z1 and the first small data is greater than the default positive correlation threshold;

S10256, the paragraph of the first small data whose positive correlation is greater than the default positive correlation threshold and the corresponding negative correlation and positive correlation are stored in the negative positive correlation paragraph set and recorded in Z2.

S1026, calculating the humor level of the corresponding paragraph based on the negative correlation level and the positive correlation level corresponding to each paragraph of the negative positive correlation paragraph set, and storing the paragraph whose humor level is higher than the default humor threshold in the humor paragraph set. .

In the present embodiment, one paragraph and a corresponding negative correlation and a positive correlation degree are obtained from the negative positive correlation paragraph set Z2, in which the negative correlation degree is 90% and the positive correlation degree is 80%. The humor level of the corresponding paragraph is calculated based on the negative correlation degree and the positive correlation degree, the paragraph in which the humor level is greater than the default humor threshold and the humor level are stored in the humor paragraph set and recorded as “Z”.

Specifically, the formula for calculating the humor level h in the relevant paragraph is as follows:
h = (pc × nc) 1/2 (6)

The quantity statistics in steps S1024 and S1025 are to adopt an existing mapreduce model and perform quick statistics for big data using hadoop or spark.

FIG. 10 is a simple sketch for obtaining a humor paragraph set, mining a paragraph set Z1 that does not match the second small data y2 (negative correlation) from the big data X, and adopting the adopted big data analysis process as Z1 = g (X , Y2), which corresponds to the negative correlation step. Mining the paragraph set Z that matches the first small data y1 from the big data X (positive correlation) and does not match the second small data y2 (negative correlation), and adopts the big data analysis process adopted as f (X, y1∪ g (X, y2)), which corresponds to the conflicting unified stairs of conflicting unified humor, that is, the stair that rises screwed to the humor identification and generation steps. The process of these two steps is not divided, but influences each other and is organized organically.

FIG. 11 is a simplified sketch of steps S1021 to S1026 described above. Big data X and second small data y2 are subjected to negative correlation analysis, and Z1 is obtained by mining from the big data. A positive correlation analysis with the small data y1 is performed, and the humor paragraph set Z is mined and acquired from the big data.

S1027, the humor paragraph with the highest humor level is selected from the humor paragraph set, and the corresponding humor paragraph is output.

This example obtains a humorous paragraph based on negative and positive correlation analysis of big data, improves the humor level of the robot, and since the negative correlation paragraph is the same as the positive correlation paragraph, the corresponding paragraph is rational and funny, Form a humor effect, screen the humor paragraph generated based on the positive and negative correlation analysis of big data through user emotion identification in the subsequent steps of Example 1, and provide better choice to the humor big data knowledge base; Further improve the robot's ability to generate humor.

Example 4:

As FIG. 12 shows, steps S101 to S106 of this embodiment are the same as those of Embodiment 1, 2, or 3, but the emotion dialogue method based on humor generation of this embodiment also includes the following:

S107, acquiring emotion feedback related information of different types of users to different responses each time based on data in the humor big data knowledge base and emotion feedback other than the default emotions of different types of users to different responses each time in the past .

The emotion feedback related information includes user information (including user type), reply content, emotion feedback content, and emotion feedback type.

As can be seen from the first embodiment, the data in the humor big data knowledge base includes the humor paragraph, the environmental information supported by the humor paragraph, the humor level of the humor paragraph, and the emotional feedback to the user's humor paragraph. Response information, environmental information includes user information, emotion feedback to the user's humor paragraph includes emotion feedback content and emotion feedback content type, emotion feedback content is the facial expression of the user, user's voice And / or user-entered text, emotion feedback type is default emotion (fun), and the data in the humor big data knowledge base is different from different types of users each time in the past Default emotional feedback to humor paragraph responses is considered to be included.

The data in the humor big data knowledge base includes default emotion feedback to the user's humor paragraph responses, but touches non-default emotions such as silence, anger, and suffering that the user has fed back to other responses Therefore, in addition to the data in the humor big data knowledge base, different types of users each time must connect emotion feedback other than the default emotion to different replies each time, so that they can get full emotion feedback related information it can.

S108, the emotion feedback related information of different types of users for different replies each time in the past is stored in emotion feedback big data.

In step S109, user information waiting for a reply is acquired, and a candidate reply set for a user waiting for a reply is acquired.

User information waiting for reply is recorded in U, candidate reply set to the user waiting for reply is recorded in P, reply content Pi is included in candidate reply set P, and i is from 1 to n , N is the number of candidate responses.

S110, obtaining expected user emotion feedback.

Anticipatory user sentiment feedback is recorded in F, and expected user sentiment feedback is obtained, for example: If the user expects to hear a reply and expects it to be fun, the expected user sentiment feedback F is fun,
Or, when the user expects to get more painful after hearing the reply, the expected user emotion feedback F is difficult.

  In S111, the emotion feedback big data is searched for all the emotion feedback records that most closely match the reply information in the user information waiting for reply and the candidate reply of the user waiting for reply.

The specific test measurement process is: Matching the user information and reply content of each emotion feedback record in the emotion feedback big data with the user information U waiting for reply and the reply content Pi of the candidate reply P to the user waiting for reply, The sentiment feedback record is obtained from the user information U waiting for a reply, the degree of coincidence with the reply content Pi of the candidate reply P to the user waiting for reply, and the sentiment feedback record corresponding to the maximum coincidence is selected therefrom, and the reply waiting Record as Ri as the emotion feedback record that most closely matches the user information U and the reply content Pi of the candidate reply P to the user waiting for reply.

S112, Emotion feedback is extracted from the emotion feedback record, and the degree of coincidence between the emotion feedback and the expected user emotion feedback is calculated and recorded in Di.

S113, based on the emotion feedback record corresponding to the greatest degree of coincidence between the emotion feedback and the expected user emotion feedback, the corresponding reply content of the corresponding emotion feedback record is acquired, and the corresponding reply content is returned to the user waiting for reply And

It can be understood that since the emotion feedback Fi in the emotion feedback Ri corresponding to the maximum matching degree Di is closest to the expected user emotion feedback F, the maximum matching degree Di corresponds to the emotion feedback recording Ri, and the corresponding emotion feedback recording Ri is Corresponding reply contents Pi are acquired, and the corresponding reply contents Pi are most likely to generate the user emotion feedback F closest to the prediction in all candidate answers.

This example shows what emotions a user will have if they respond differently to the same words of the user through data in the humor big data knowledge base and emotion feedback other than the default emotion to different responses in the past. It predicts what will happen, and then selects a reply that will give the user a specific feeling based on the prediction result.

Those skilled in the art can understand that all or some of the steps of the methods of the above embodiments are realized by a program indicating related hardware, and the corresponding program can be stored in a computer-readable storage medium. The storage medium is, for example, a ROM / RAM, a magnetic disk or a disk.

Example 5:

As shown in FIG. 13, this embodiment provides a robot system based on humor generation, which includes a user data acquisition module 1301, a humor paragraph generation output module 1302, an emotion feedback monitoring module 1303, a first determination module 1304, Including a dual decision module 1305 and a humor big data knowledge base acquisition module 1306, the specific functions of each module are as follows:

The user data acquisition module 1301 acquires user data.

The humor paragraph generation / output module 1302 generates and outputs a humor paragraph based on user data, and the humor paragraph is generated by an existing humor generation method,

The emotion feedback monitoring module 1303 monitors the user's emotion feedback to the output humor paragraph,

When the user's emotion feedback is identified as the default emotion, the first determination module 1304 determines that the output humor paragraph has a humor characteristic, and determines the default emotion level based on the identification result of the user's emotion feedback. The larger the, the higher the humor level of the output humor paragraph,

When the user's emotion feedback is identified as an emotion other than the default emotion, the second determination module 1305 determines that the output humor paragraph does not have humor characteristics,

  The humor big data knowledge base acquisition module 1306 humor big data knowledge of humor paragraph having humor characteristics, environmental information corresponding to the humor paragraph, humor level of the humor paragraph, and user emotion feedback to the humor paragraph. Save to base.

Example 6:

The other modules of the present embodiment are the same as those of the first embodiment, and a different part is in the implementation of the humor paragraph generation / output module 1302. As shown in FIG. 14, the humor paragraph generation / output module 1302 includes a small data extraction unit 13021, Each unit includes a big data acquisition unit 13022, a first correlation data set acquisition unit 13023, a positive correlation paragraph set acquisition unit 13024, a positive / negative correlation paragraph set acquisition unit 13025, a humor paragraph set acquisition unit 13026, and a humor paragraph output unit 13027. Typical features are as follows:

The small data extraction unit 13021 extracts some data from the user data as first small data, and extracts some other data as second small data.

The big data acquisition unit 13022 acquires big data.

The first correlation data set acquisition unit 13023 obtains data related to the first small data from the big data and cleans it to obtain a first correlation data set.

The positive correlation paragraph set acquisition unit 13024 mines a paragraph whose positive correlation with the first small data is greater than the default positive correlation threshold from the first correlation data set, and obtains the mined paragraph and its positive correlation. The specific process is stored in the positive correlation paragraph set, and the specific process can be referred to the second embodiment.

The positive / negative correlation paragraph set acquisition unit 13025 mines a paragraph in which the negative correlation with the second small data is greater than the default negative correlation threshold from the positive correlation paragraph set, and the mined paragraph and its positive correlation and negative correlation The degree of correlation is stored in a set of positive and negative correlation paragraphs, and the specific process can be referred to Example 2.

The humor paragraph set acquisition unit 13026 calculates the humor level of the paragraph based on the positive correlation and the negative correlation corresponding to each paragraph of the positive / negative correlation paragraph set, and the humor level is greater than the default humor threshold. It is stored in a paragraph set, and the specific process can be referred to Example 2.

The humor paragraph output unit 13027 selects the humor paragraph having the highest humor level from the humor paragraph set, and outputs the corresponding humor paragraph.

Example 7:

Other modules of the present embodiment are the same as those of the first embodiment, and a different part is in the realization of the humor paragraph generation / output module 1302. As shown in FIG. 15, the humor paragraph generation / output module 1302 includes a small data extraction unit 13021, Each unit includes a big data acquisition unit 13022, a first correlation data set acquisition unit 13023, a negative correlation paragraph set acquisition unit 13024, a negative positive correlation paragraph set acquisition unit 13025, a humor paragraph set acquisition unit 13026, and a humor paragraph output unit 13027. Specific features are as follows:

The small data extraction unit 13021 extracts some data from the user data as first small data, and extracts some other data as second small data.

The big data acquisition unit 13022 acquires big data.

The first correlation data set acquisition unit 13023 obtains data related to the first small data from the big data and cleans it to obtain a first correlation data set.

The negative correlation paragraph set acquisition unit 13024 mines a paragraph in which the negative correlation with the second small data is larger than the default negative correlation threshold from the first correlation data set, and determines the mined paragraph and its negative correlation. It is stored in the negative correlation paragraph set, and the specific process can be referred to Example 3.

The negative positive correlation paragraph set acquisition unit 13025 mines a paragraph in which the positive correlation with the first small data is greater than the default positive correlation threshold from the negative correlation paragraph set, the mined paragraph and its negative correlation The degree of positive correlation is stored in a set of negative and positive correlation paragraphs, and the specific process can be referred to Example 3.

The humor paragraph set acquisition unit 13026 calculates the humor level of the corresponding paragraph based on the negative correlation and the positive correlation corresponding to each paragraph of the negative positive correlation paragraph set, and the humor level is greater than the default humor threshold. Is stored in a humor paragraph set, and the specific process can be referred to Example 3.

  The humor paragraph output unit 13027 selects the humor paragraph having the highest humor level from the humor paragraph set, and outputs the corresponding humor paragraph.

Example 8:

As shown in FIG. 16, the user data acquisition module 1301, the humor paragraph generation output module 1302, the emotion feedback monitoring module 1303, the first determination module 1304, the second determination module 1305, and the humor big data knowledge base acquisition module 1306 of the present embodiment. Is the same as in the fifth, sixth or seventh embodiment, but the robot system based on the generation of humor in the present embodiment includes an emotion feedback related information acquisition module 1307, an emotion feedback big data acquisition module 1308, a candidate answer set acquisition module 1309, an expected user An emotion feedback acquisition module 1310, an emotion feedback record search module 1311, a coincidence calculation module 1312, and a reply content acquisition module 1313; Specific functions of these modules are as follows:

The emotion feedback related information acquisition module 1307 is configured to use different types of users for different responses each time in the past based on the data in the humor big data knowledge base and emotion feedback other than the default emotions of different types of users for different responses in the past. The emotion feedback related information, the emotion feedback related information includes user information, reply content, emotion feedback content and emotion feedback type,

The emotion feedback big data acquisition module 1308 stores emotion feedback related information of different types of users to different responses each time in the past in emotion feedback big data,

The candidate answer set acquisition module 1309 acquires user information waiting for a reply, acquires a candidate reply set for a user waiting for a reply,

The expected user emotion feedback acquisition module 1310 acquires expected user emotion feedback,

The emotion feedback record retrieval module 1311 retrieves all emotion feedback records that most closely match the reply information in the candidate reply set to the user information waiting for reply and the user waiting for reply from the emotion feedback big data. like:

User information for each emotional feedback record in emotional feedback big data, user response information is matched to the user information waiting for reply, and the answer content in the candidate answer set for the user waiting for reply, each emotion feedback record and user information waiting for reply , Get the degree of match with the candidate reply set for the user waiting for reply, select the corresponding emotion feedback record with the largest match from among them, select the user information waiting for reply and the candidate reply to the user waiting for reply The emotional feedback record that best matches the set.

The coincidence calculation module 1312 extracts emotion feedback from the emotion feedback record, and calculates the coincidence between the emotion feedback and the expected user emotion feedback.

The reply content acquisition module 1313 acquires the reply content corresponding to the corresponding emotion feedback record based on the emotion feedback record corresponding to the greatest degree of coincidence between the emotion feedback and the expected user emotion feedback, and returns the corresponding reply content. Answer to the waiting user.

What should be explained here is that the robot system provided in the embodiment is described by taking an example according to the division of each functional module, and in an actual application, the function is distributed to each functional module according to demand. Completion, that is, the internal structure is divided into functional modules, and all or some of the above functions are completed.

It can be understood that the terms “first” and “second” used in the robot systems provided in the above embodiments can describe various modules, but these modules are not subject to the terminology restrictions. is there. These terms are only used to distinguish one module from another. For example, if not departing from the scope of the present invention, the first judgment module can be called the second judgment module, and similarly, the second judgment module can be called the first judgment module, and the first judgment module and the second judgment module are Either is a decision module, but not the same decision module.

What has been described above is merely a specific implementation method of the present invention, but the protection scope of the present invention is not limited thereto. Changes and replacements that have been conceived not through all creative labor are covered by the protection scope of the present invention. Therefore, the protection scope of the present invention is based on the protection scope in which the rights request is limited.

Claims (2)

Features of the emotion interaction method based on humor generation include the following when the method is applied to a robot:
Get user data,
Generate and output humor paragraphs based on user data,
Monitor the user's emotional feedback to the output humor paragraph,
When the user's emotion feedback is identified as the default emotion, the output humor paragraph is judged to have humor characteristics, and the higher the default emotion level in the user's emotion feedback identification result, the greater the output humor paragraph. The level of humor increases,
If the user ’s emotional feedback is identified as an emotion other than the default emotion, the output humor paragraph will refrain from having humor characteristics,
Humor paragraph with humor characteristics, environmental information corresponding to the relevant humor paragraph, humor level of the relevant humor paragraph and user emotion feedback for the relevant humor paragraph are stored in the humor big data knowledge base,
Generate and output a humor paragraph based on the user's data, specifically including: Extracting some data from the user's data as first small data and another data as second small data Extract and
Get big data,
Obtain the data related to the first small data from the big data and clean it to make the first correlation data set,
Mining a paragraph in which the positive correlation with the first small data from the first correlation data set is greater than the default positive correlation threshold, storing the mined paragraph and its positive correlation in a positive correlation paragraph set,
Mining a paragraph whose negative correlation with the second small data is greater than the default negative correlation threshold from the positive correlation paragraph set and storing the mined paragraph and its positive and negative correlation degrees in the positive and negative correlation paragraph set ,
Calculate the humor level of the corresponding paragraph based on the positive correlation and negative correlation corresponding to each paragraph of the positive / negative correlation paragraph set, and save the paragraph whose humor level is higher than the default humor threshold in the humor paragraph set,
Select the humor paragraph with the highest humor level from the set of humor paragraphs, output the relevant humor paragraph,
Generate and output a humor paragraph based on the user's data, specifically including: Extracting some data from the user's data as first small data and another data as second small data Extract and
Get big data,
Obtain the data related to the first small data from the big data and clean it to make the first correlation data set,
Mining a paragraph in which the negative correlation with the second small data from the first correlation data set is greater than the default negative correlation threshold, storing the mined paragraph and its negative correlation in a negative correlation paragraph set,
Mines the negative correlation paragraph set that has a positive correlation with the first small data that is greater than the default negative correlation threshold, and saves the mined paragraph and its negative and positive correlation degrees in the negative positive correlation paragraph set. And
Calculate the humor level of the corresponding paragraph based on the negative correlation and the positive correlation corresponding to each paragraph of the negative positive correlation paragraph set, and save the paragraph whose humor level is higher than the default humor threshold in the humor paragraph set,
Select the humor paragraph with the highest humor level from the set of humor paragraphs, output the relevant humor paragraph,
The method includes:
Based on the data in the humor big data knowledge base and the emotion feedback other than the default emotion of different types of users to different responses each time in the past, the emotion feedback related information of different types of users to different responses in the past is obtained and The emotion feedback related information includes user information, reply content, emotion feedback content and emotion feedback type,
Store emotional feedback related information of different types of users for different responses in the past in emotion feedback big data,
Get user information waiting for a reply, get a candidate reply set for users waiting for a reply,
Get anticipatory user sentiment feedback,
Find all sentiment feedback records that most closely match the sentiment feedback big data to the user information awaiting reply and the reply content in the candidate reply set to the waiting reply user,
Extract emotional feedback from emotional feedback records, calculate the degree of agreement between emotional feedback and expected user emotional feedback,
Based on the emotion feedback record corresponding to the greatest degree of agreement between the emotion feedback and the expected user emotion feedback, the corresponding response content of the corresponding emotion feedback record is obtained, and the corresponding response content is returned to the user waiting for the response.
From the sentiment feedback big data, search for all sentiment feedback records that most closely match the user information and the contents of the candidate answer set, specifically:
User information for each emotional feedback record in emotional feedback big data, match the response content to the user information waiting for the response, the response content in the candidate answer set for the user waiting for the response, each emotion feedback record and the user information waiting for the response , Get the degree of match with the candidate reply set for the user waiting for reply, select the corresponding emotion feedback record with the largest match from among them, select the user information waiting for reply and the candidate reply to the user waiting for reply As the emotional feedback record that best matches the set, Emotion interaction method and robot system based on humor generation, said system comprising user data acquisition module, humor paragraph generation output module, emotion feedback monitoring module, first determination module, second determination module and humor big data knowledge base acquisition module Including
The user data acquisition module acquires user data,
The humor paragraph generation / output module generates and outputs a humor paragraph according to user data,
The emotion feedback monitoring module monitors the user's emotion feedback to the output humor paragraph,
When the user's emotional feedback is identified as the default emotion, the first determination module determines that the output humor paragraph has humor characteristics, and the user's emotional feedback identification results in a high default emotion level. The higher the humor level of the output humor paragraph,
The second determination module refuses to say that the output humor paragraph has no humor characteristics when the user's emotion feedback is identified as an emotion other than the default emotion,
The humor big data knowledge base acquisition module includes humor paragraphs having humor characteristics, environmental information corresponding to the humor paragraphs, humor levels of the humor paragraphs, and user emotion feedback to the humor paragraphs. Save to
The humor paragraph generation output module includes a small data extraction unit, a big data acquisition unit, a first correlation data set acquisition unit, a positive correlation paragraph set acquisition unit, a positive / negative correlation paragraph set acquisition unit, a humor paragraph set acquisition unit, and a humor paragraph output unit. Including
The small data extraction unit extracts a part of the data from the user data as the first small data and the other part as the second small data.
The big data acquisition unit acquires big data,
The first correlation data set acquisition unit obtains data related to the first small data from the big data and cleans it to obtain a first correlation data set,
The positive correlation paragraph set acquisition unit mines the paragraph in which the positive correlation with the first small data from the first correlation data set is greater than the default positive correlation threshold, and corrects the mined paragraph and its positive correlation. Save in a set of correlated paragraphs,
The positive / negative correlation paragraph set acquisition unit mines a paragraph whose negative correlation with the second small data is larger than the default negative correlation threshold from the positive correlation paragraph set, and the mined paragraph and its positive correlation and negative correlation Store degrees in a positive and negative correlation paragraph set,
The humor paragraph set acquisition unit calculates the humor level of the corresponding paragraph based on the positive correlation and the negative correlation corresponding to each paragraph of the positive / negative correlation paragraph set, and the humor level is greater than the default humor threshold. Save it in a collective,
The humor paragraph output unit selects the humor paragraph with the highest humor level from the humor paragraph set, outputs the corresponding humor paragraph,
The humor paragraph generation output module includes a small data extraction unit, a big data acquisition unit, a first correlation data set acquisition unit, a negative correlation paragraph set acquisition unit, a negative positive correlation paragraph set acquisition unit, a humor paragraph set acquisition unit, and a humor paragraph output Including units,
The small data extraction unit extracts some data from the user's data as first small data, and extracts some data as second small data,
The big data acquisition unit acquires big data,
The first correlation data set acquisition unit obtains data related to the first small data from the big data and cleans it to obtain a first correlation data set,
The negative correlation paragraph set acquisition unit mines a paragraph in which the negative correlation with the second small data from the first correlation data set is greater than the default negative correlation threshold, and negatives the mined paragraph and its negative correlation. Save in a set of correlated paragraphs,
The negative positive correlation paragraph set acquisition unit mines a paragraph in which the positive correlation degree with the first small data is larger than the default positive correlation degree threshold from the negative correlation paragraph set, the mined paragraph and its negative correlation degree and the positive correlation degree. Save the degree of correlation in the negative positive correlation paragraph set,
The humor paragraph set acquisition unit calculates a humor level of a corresponding paragraph based on a negative correlation level and a positive correlation level corresponding to each paragraph of the negative positive correlation paragraph set, and calculates a humor level greater than a default humor threshold. Save it in a humor paragraph set,
The humor paragraph output unit selects the humor paragraph with the highest humor level from the humor paragraph set, outputs the corresponding humor paragraph,
The system includes an emotion feedback related information acquisition module, an emotion feedback big data acquisition module, a candidate reply set acquisition module, an expected user emotion feedback acquisition module, an emotion feedback record search module, a matching degree calculation module, and an answer content acquisition module. Feedback related information acquisition module is based on the data in the humor big data knowledge base and emotion feedback other than the default emotion of different types of users to different responses each time in the past, different types of user emotion feedback to different responses in the past The related information is acquired, and the emotion feedback related information includes user information, reply content, emotion feedback content and emotion feedback type.
The emotion feedback big data acquisition module stores emotion feedback related information of different types of users to different responses each time in the past in emotion feedback big data,
The candidate answer set acquisition module acquires user information waiting for a reply, acquires a candidate answer set for a user waiting for a reply,
The anticipation user emotion feedback acquisition module acquires anticipation user emotion feedback,
The sentiment feedback record search module searches all sentiment feedback records that most closely match the reply information in the candidate answer set to the waiting user information and the waiting reply user information from the sentiment feedback big data,
The coincidence calculation module extracts emotion feedback from the emotion feedback record, calculates the agreement between the emotion feedback and the expected user emotion feedback,
The reply content acquisition module acquires the reply content corresponding to the corresponding emotion feedback record based on the emotion feedback record corresponding to the greatest degree of coincidence between the emotion feedback and the expected user emotion feedback, and waits for the corresponding reply content As a reply to the user
The emotion feedback record search module is specifically:
User information for each emotional feedback record in emotional feedback big data, match the response content to the user information waiting for the response, the response content in the candidate answer set for the user waiting for the response, each emotion feedback record and the user information waiting for the response , Get the degree of match with the candidate reply set for the user waiting for reply, select the corresponding emotion feedback record with the largest match from among them, select the user information waiting for reply and the candidate reply to the user waiting for reply The emotional feedback record that best matches the set.

Images