Information interaction method, device, equipment and readable storage medium

Abstract

The present invention discloses an information interaction method, device, equipment and readable storage medium, including receiving first input information of a target object; if first attribute information corresponding to the first input information does not meet a preset condition, generating first output information according to the first attribute information; the first attribute information is part of the first input information and corresponds to the intention information of the first input information; if the first attribute information and the second attribute information corresponding to the second input information meet a preset condition, generating second output information according to the first attribute information and the second attribute information; the second attribute information is at least part of the second input information and corresponds to the intention information of the second input information. The attribute information of the target object scattered in multiple rounds of dialogues is extracted multiple times by information interaction, and the intention is understood according to all the attribute information extracted in the multiple rounds of dialogues, thereby improving the accuracy of the intelligent customer service feedback information and also improving the user experience.

Description

Technical Field

[0001] The present invention relates to the field of information interaction technology, and in particular to an information interaction method, device, equipment and readable storage medium. Background Art

[0002] When solving user problems, the current intelligent customer service system identifies user intent and extracts slots, and then pushes corresponding answers based on intent and slots. Currently, the commonly used methods for intent recognition are text classification and text matching methods, which map user questions to intent.

[0003] In actual application, since users express themselves more casually, information is usually input intermittently in multiple rounds, resulting in the key information needed to determine the intent being scattered in different rounds of dialogue, which brings great difficulties to intent understanding. Summary of the invention

[0004] The embodiments of the present invention provide an information interaction method, apparatus, device and readable storage medium, which have the technical effects of improving reply accuracy and improving user experience.

[0005] On the one hand, the present invention provides an information interaction method, which includes: receiving first input information of a target object; if first attribute information corresponding to the first input information does not meet a preset condition, generating first output information based on the first attribute information, and instructing the target object to feedback second input information based on the first output information; wherein the first attribute information is at least part of the first input information and corresponds to intention information of the first input information; if the first attribute information and the second attribute information corresponding to the second input information meet the preset condition, generating second output information based on the first attribute information and the second attribute information and feeding it back to the target object; wherein the second attribute information is at least part of the second input information and corresponds to the intention information of the second input information, and the second output information includes at least information generated based on the first attribute information and the second attribute information.

[0006] In one possible implementation, the first attribute information is extracted based on a first preset information set, wherein the first preset information set includes an attribute type corresponding to the attribute information; based on the intention information in the first input information, the corresponding first attribute information is extracted from the first input information, wherein the extracted first attribute information corresponds to the attribute type in the first preset information set.

[0007] In one possible implementation, the first attribute information corresponding to the first input information does not meet a preset condition, including at least one of the following: the amount of the first attribute information is lower than a first threshold; the first attribute information does not correspond to an attribute type in the first preset information set.

[0008] In one possible implementation manner, generating the second output information based on the first attribute information and the second attribute information includes: determining the intent correlation between the first input information and the second input information; determining a corresponding intermediate preset information set based on the intent information of the second input information; if it is determined that the intent correlation does not exceed a preset threshold, generating a second preset information set based on the first preset information set and the intermediate preset information set; if it is determined that the intent correlation exceeds the preset threshold, using the intermediate preset information set as the second preset information set; and generating the second output information based on the second preset information set.

[0009] In one possible implementation manner, the intermediate preset information set includes multiple attribute types; accordingly, generating the second preset information set based on the first preset information set and the intermediate preset information set includes: storing the second attribute information in the intermediate preset information set according to the attribute type; replacing the first attribute information in the first preset information set according to the second attribute information in the intermediate preset information set, or adding the second attribute information to the first preset information set.

[0010] In one possible implementation, generating the second output information according to the first attribute information and the second attribute information includes: querying corresponding intent information from an information library based on the first attribute information and the second attribute information; and generating the second output information according to the queried intent information.

[0011] In one possible implementation manner, the first attribute information / second attribute information is obtained in the following manner: by understanding the intent of the first input information / second input information; or by sequence labeling or multi-label classification.

[0012] On the other hand, the present invention provides an information interaction device, which includes: an information receiving module, which is used to receive first input information of a target object; a first output module, which is used to generate first output information based on the first attribute information if the first attribute information corresponding to the first input information does not meet the preset conditions, and instruct the target object to feedback second input information based on the first output information; wherein the first attribute information is at least part of the first input information and corresponds to the intention information of the first input information; a second output module, which is used to generate second output information based on the first attribute information and the second attribute information and feedback it to the target object if the second attribute information corresponding to the first attribute information and the second input information meets the preset conditions; wherein the second attribute information is at least part of the second input information and corresponds to the intention information of the second input information, and the second output information includes at least information generated based on the first attribute information and the second attribute information.

[0013] Another aspect of the present invention provides an information interaction device, the device comprising a memory and a processor,

[0014] The memory is used to store instructions, and the instructions are used to control the processor to operate to implement any of the above-mentioned information interaction methods when executed.

[0015] Another aspect of the present invention provides a computer-readable storage medium, which includes a group of computer-executable instructions, and when the instructions are executed, is used to execute any of the above-mentioned information interaction methods.

[0016] In an embodiment of the present invention, attribute information of the target object scattered in multiple rounds of conversations is extracted multiple times by means of information interaction, and intention understanding is performed based on all attribute information extracted in the multiple rounds of conversations, thereby improving the accuracy of intelligent customer service feedback information and also improving user experience. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] The above and other objects, features and advantages of the exemplary embodiments of the present invention will become readily understood by reading the detailed description below with reference to the accompanying drawings. In the accompanying drawings, several embodiments of the present invention are shown in an exemplary and non-limiting manner, in which:

[0018] In the drawings, the same or corresponding reference numerals represent the same or corresponding parts.

[0019] Figure 1 A schematic diagram of an implementation flow of an information interaction method according to an embodiment of the present invention;

[0020] Figure 2 A schematic diagram of human-computer interaction in an information interaction method according to an embodiment of the present invention;

[0021] Figure 3 A schematic diagram of replacement / update between preset information sets in an information interaction method according to an embodiment of the present invention;

[0022] Figure 4 The figure is a schematic diagram of the structure of an information interaction device according to an embodiment of the present invention. DETAILED DESCRIPTION

[0023] In order to make the purpose, features and advantages of the present invention more obvious and easy to understand, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present invention.

[0024] like Figure 1 As shown, the present invention provides an information interaction method on one hand, the method comprising:

[0025] Step 101, receiving first input information of a target object;

[0026] Step 102: if the first attribute information corresponding to the first input information does not meet the preset condition, generate first output information according to the first attribute information, and instruct the target object to feedback second input information according to the first output information;

[0027] The first attribute information is at least part of the first input information and corresponds to the intention information of the first input information;

[0028] Step 103: if the first attribute information and the second attribute information corresponding to the second input information meet a preset condition, second output information is generated according to the first attribute information and the second attribute information and fed back to the target object;

[0029] The second attribute information is at least part of the second input information and corresponds to the intention information of the second input information, and the second output information at least includes information generated according to the first attribute information and the second attribute information.

[0030] In this embodiment, the information interaction method can be applied to a server side, a cloud side, or a user terminal. For the sake of clarity, application to a server side is taken as an example.

[0031] In step 101, the target object includes a person, an animal, or a robot with voice function, a virtual character, etc. The first input information includes text information, voice information, etc. The first input information can be transmitted by the target object through a hardware device such as a keyboard and a mouse to the server side, or can be transmitted to the server side through a wireless transmission method such as Bluetooth and the Internet.

[0032] In step 102, after receiving the first input information, the server performs intent recognition on the first input information to obtain intent information. If the received first input information is voice information, the existing voice recognition technology is used to process the text information and then perform intent recognition. Intent recognition is to identify the intention or purpose of the input information. For example, if the first input information is "My mobile phone has been unable to automatically connect to the Bluetooth device via Bluetooth, and I can only use WiFi connection all the time, which makes me very distressed", its intent information is "Bluetooth cannot connect automatically", and the intent recognition method can be specifically obtained by identifying the input information through the intent recognition algorithm in the existing natural language understanding.

[0033] Based on the intent information, select relevant characters, words, sentences and other partial information from the first input information as the first attribute information. When selecting, you can traverse and determine whether the characters, words and sentences in the first input information are associated with the intent information. The associated judgment method can be to input the characters, words, sentences and intent information into the existing knowledge graph to determine whether they belong to the same field. If they belong to the same field, it is determined that the characters, words and sentences are associated with the intent information. Taking the first input information "My mobile phone's WiFi has a problem" as an example, the first attribute information selected from it may be "mobile phone" or "WiFi".

[0034] Determine whether the first attribute information meets the preset condition, wherein the preset condition is a pre-set rule, such as determining whether the amount of the selected first attribute information is sufficient. If it is determined that the first attribute does not meet the preset condition, the server may not be able to accurately reply to the target object's desired result based on the currently selected first attribute information, then the server selects the corresponding reply statement from the pre-stored reply template as the first output information based on the first attribute information, wherein the first output information is preferably information in the form of a rhetorical question, and continues to take the first input information as "My mobile phone WiFi has a problem" as an example, the server queries the first output information from the pre-stored reply template based on the first attribute information - "mobile phone", "WiFi", such as "What's the problem with mobile phone WiFi?", and sends the first output information to the target object, and instructs the target object to feedback the second input information based on the first output information, and the instruction method can be to add fixed words such as "Please answer" on the basis of the first output information to prompt the user.

[0035] In step 103, after receiving the second input information of the target object, the server performs intent recognition on the second input information to obtain corresponding intent information, and then selects relevant characters, words, sentences and other partial information from the second input information as second attribute information based on the intent information of the second input information. The method for recognizing the intent information of the second input information and the method for selecting the second attribute information are the same as the processing method for the first input information, and will not be elaborated here.

[0036] Determine whether the first attribute information and the second attribute information meet the preset conditions, for example, determine whether the number of the selected first attribute information and the second attribute information is sufficient. If it is determined that the first attribute information and the second attribute information meet the above preset conditions, the first attribute information and the second attribute information can be integrated, wherein the integration method can be the mutual superposition of the attribute information, or the replacement of the attribute information, and determine whether the integrated attribute information still has corresponding intention information. The method for determining whether there is corresponding intention information can be to establish a mapping relationship between multiple attribute information and intention information in advance and store them in the information library. When the integrated attribute information is queried in the information library, it is determined that the integrated attribute information has corresponding intention information. After the integrated attribute information is queried, the corresponding intention information is extracted from the information library, and the reply information is generated according to the extracted intention information, and is used as at least a part of the second output information, wherein the specific method for generating the reply information can be to input the extracted intention information into the dialogue system, and generate the reply information through the dialogue system. Of course, in addition to the reply information, the second output information can also include fixed speech information.

[0037] If the server determines that the first attribute information and the second attribute information still do not meet the preset conditions, the third output information is generated based on the first attribute information and the second attribute information and fed back to the target object, and the above steps are repeated, wherein the generation process of the third output information is the same as the above-mentioned method of generating the first output information, and will not be repeated here. Furthermore, during the feedback process, the server can also set the number of feedbacks on the server side, and the number of feedbacks is accumulated every time the attribute information is determined to not meet the preset conditions. When the accumulated number exceeds the number of feedbacks on the server side, the server side can end the current dialogue and switch to manual service.

[0038] Therefore, the attribute information of the target object scattered in multiple rounds of conversations can be extracted multiple times through information interaction, and the intention can be understood based on all the attribute information extracted in multiple rounds of conversations, thereby improving the accuracy of intelligent customer service feedback information and improving user experience.

[0039] In one possible implementation, the first attribute information is extracted based on a first preset information set, wherein the first preset information set includes an attribute type corresponding to the attribute information;

[0040] Based on the intention information in the first input information, corresponding first attribute information is extracted from the first input information, wherein the extracted first attribute information corresponds to an attribute type in the first preset information set.

[0041] In this embodiment, after extracting the first attribute information, the server side stores the extracted first attribute information in a first preset information set, wherein the first preset information set is a pre-set data set and a mapping relationship is pre-established with the intent information of the first input information. After identifying the intent information in the first input information, the corresponding first preset information set can be obtained according to the mapping relationship.

[0042] The first preset information set includes at least one attribute type, such as subject, predicate and object. According to the attribute type in the first preset information set, the first attribute information is extracted from the first input information. Taking the preset information set including three attribute types of subject, predicate and object, and the input information being "the mobile phone cannot automatically connect to WiFi" as an example, when extracting the attribute information, the attribute information extracted from the input information is "mobile phone", "cannot automatically connect" and "WiFi". The extraction method can specifically be to obtain the type of each word in the first input information during the process of understanding the intent of the first input information through the existing intent understanding module, match the type of each word with the attribute type, extract the successfully matched words and use them as the first attribute information. The specific extraction method can also be obtained through sequence annotation or multi-label classification, that is, a large number of words are established in advance, and each word is sequence-annotated or labeled, and the words in the first input information are matched with the words after sequence annotation or multi-label classification. If the match is successful, the same word type is set for the words in the first input information, and then the words that have been set to the same word type are matched with the attribute type of the first preset information set, and the successfully matched words are extracted and used as the first attribute information.

[0043] In one possible implementation, the first attribute information corresponding to the first input information does not satisfy a preset condition, including at least one of the following:

[0044] The amount of the first attribute information is lower than a first threshold;

[0045] The first attribute information does not correspond to an attribute type in the first preset information set.

[0046] In this embodiment, the manner in which the first attribute information does not meet the preset condition mainly includes:

[0047] 1. The amount of the first attribute information is lower than the first threshold, where the first threshold can be set according to actual application requirements. When the amount of the first attribute information is lower than the first threshold, it may be difficult to query accurate reply information to the target object with the current amount of the first attribute information, which will lead to a decrease in user experience.

[0048] 2. The first attribute information does not correspond to the attribute type in the first preset information set. In this case, there will be no corresponding attribute information for one or more attribute types in the first preset information set, which may make it difficult to query accurate reply information for the target object based on the first attribute information in the current first preset information set, thereby reducing the user experience.

[0049] In one possible implementation, generating the second output information according to the first attribute information and the second attribute information includes:

[0050] determining an intent relevance between the first input information and the second input information;

[0051] Determining a corresponding intermediate preset information set based on the intention information of the second input information;

[0052] If it is determined that the intention relevance does not exceed the preset threshold, generating a second preset information set based on the first preset information set and the intermediate preset information set;

[0053] If the determination intention relevance exceeds a preset threshold, the intermediate preset information set is used as the second preset information set;

[0054] Generate second output information according to the second preset information set.

[0055] In this embodiment, the intent relevance is the similarity between the intent information representing the first input information and the intent information representing the second input information. The similarity calculation method can obtain the first input information and the second input information by using them as inputs of the text similarity algorithm in the existing natural language processing.

[0056] At the same time, based on the intent information of the second input information, the corresponding intermediate preset information set is determined, wherein the intermediate preset information set is also a preset data set like the first preset information set, and also establishes a mapping relationship with the intent information of the second input information in advance, and also includes at least one attribute type. The process of determining the intermediate preset information set is specifically: querying the corresponding intermediate preset information set through the mapping relationship of the intent information of the second input information, and extracting the second attribute information from the second input information according to the attribute type in the intermediate preset information set, wherein the extraction method is the same as the above-mentioned method of extracting the first attribute information, and will not be repeated here. The extracted second attribute information is stored in the intermediate preset information set according to the attribute type.

[0057] Determine whether the intention relevance exceeds a preset threshold, where the preset threshold is a fixed value and can be adjusted according to actual needs.

[0058] If it is determined that the intention relevance exceeds the preset threshold, it means that the intention relevance between the first input information and the second input information is relatively high, that is, the first input information and the second input information both describe the same thing. Then, the first preset information set and the intermediate preset information set are added or replaced according to the attribute type. The addition process is as follows: If there is a second attribute information of a certain attribute type in the intermediate preset information set, but there is no attribute information of the same attribute type in the first preset information set, then add the second attribute information to the first preset information set. For example, the first preset information set is <mobile phone, None, WiFi>, and the intermediate preset information set is <None, cannot automatically connect, None>. After addition, the final obtained second preset information set is <mobile phone, cannot automatically connect, WiFi>. The replacement process is as follows: If there is a second attribute information of a certain attribute type in the intermediate preset information set, and there is also attribute information of the same attribute type in the first preset information set, then replace the attribute information in the first preset information set with the second attribute information. For example, the first preset information set is <mobile phone, cannot automatically connect, WiFi>, and the intermediate preset information set is <None, None, Bluetooth>. After replacement, the final obtained second preset information set is <mobile phone, cannot automatically connect, Bluetooth>.

[0059] If it is determined that the intention relevance does not exceed the preset threshold, it means that the intention relevance between the first input information and the second input information is relatively low, that is, the first input information and the second input information are two completely unrelated things. Then, discard the original first preset information set and directly use the intermediate preset information set as the second preset information set.

[0060] Generate second output information according to the second preset information set; specifically, determine whether there is a corresponding intention information in the information library for the attribute information in the current second preset information set. If it is determined that there is a corresponding intention information in the information library, then generate a reply message according to the intention information and use it as at least a part of the second output information. Of course, in addition to the reply message, the second output information can also include fixed话术 information, etc.

[0061] Combined Figure 2 and Figure 3 as shown, elaborate on the entire process:

[0062] After the server receives the first input information "There is a problem with the WiFi connection on my mobile phone" sent by the user, it performs intent recognition on the first input information, generates intent information, and queries the first preset information set according to the intent information. The first attribute information "mobile phone" and "WiFi" are queried from the first input information according to the attribute types in the first preset information set. It can be seen that there is an attribute type in the current preset information set that does not contain the first attribute information. Therefore, the first output information in the form of a rhetorical question "What's the problem with the mobile phone WiFi?" is sent to the user based on the existing first attribute type.

[0063] The server receives the second input information sent by the user according to the first output information, queries the intermediate preset information set according to the intent of the second input information, and queries the second attribute information "cannot connect automatically" from the second input information according to the attribute types in the intermediate preset information set and stores it in the intermediate preset information set, obtaining <None, cannot connect automatically, None>; calculates the intent relevance between the intent of the first input information and the intent of the second input information (combined with Figure 3 ). It can be seen that the intent relevance between the two is relatively high, and the second attribute information "cannot connect automatically" is the attribute type missing from the first preset information set. Therefore, the first preset information set and the intermediate preset information set are added to obtain the second preset information set <mobile phone, cannot connect automatically, WI-FI>, that is, the current preset information set. At this time, the current preset information set meets the preset conditions, so the corresponding intent information "WiFi cannot connect automatically" is queried from the current preset information set, and the second output information containing the reply information "It seems that your WiFi cannot connect automatically, and the solution is..." is fed back to the user according to the queried intent information. If the current preset information set still does not meet the preset conditions at this time, corresponding rhetorical questions are generated according to the attribute information in the second preset information set, and the above steps are repeated.

[0064] At this time, the server receives the third input information "Does the X model mobile phone support Volte?" sent by the user, and intends to understand the intent information "The X model mobile phone supports Volte" of the third input information. Calculates the intent relevance between "The X model mobile phone supports Volte" and "WiFi cannot connect automatically". It can be seen that the intent relevance between the two is relatively small. Therefore, "The X model mobile phone supports Volte" is used as the current intent, and the current preset information set <X model mobile phone, support, Volte> is updated according to the current intent. The attribute information in the current current preset information set also meets the preset conditions. Therefore, the intent "support Volte" corresponding to <mobile phone, support, Volte> is queried, and the reply information "The X model mobile phone supports Volte" is fed back according to the queried intent "support Volte".

[0065] As Figure 4As shown, another aspect of the present invention provides an information interaction device, the device comprising:

[0066] The information receiving module 201 is used to receive first input information of a target object;

[0067] A first output module 202, configured to generate first output information according to the first attribute information if the first attribute information corresponding to the first input information does not meet a preset condition, and instruct the target object to feedback second input information according to the first output information;

[0068] The first attribute information is at least part of the first input information and corresponds to the intention information of the first input information;

[0069] The second output module 203 is used to generate second output information according to the first attribute information and the second attribute information and feed it back to the target object if the second attribute information corresponding to the first attribute information and the second input information meets a preset condition;

[0070] The second attribute information is at least part of the second input information and corresponds to the intention information of the second input information, and the second output information at least includes information generated according to the first attribute information and the second attribute information.

[0071] In this embodiment, the information interaction method can be applied to a server side, a cloud side, or a user terminal. For the sake of clarity, application to a server side is taken as an example.

[0072] In the information receiving module 201, the target object includes a person, an animal, or a robot with voice function, a virtual character, etc. The first input information includes text information, voice information, etc. The first input information can be transmitted by the target object through a hardware device such as a keyboard and a mouse to the server side, or can be transmitted to the server side through a wireless transmission method such as Bluetooth and the Internet.

[0073] In the first output module 202, after receiving the first input information, the server performs intent recognition on the first input information to obtain intent information. If the received first input information is voice information, the existing voice recognition technology is used to process the text information and then perform intent recognition. Intent recognition is to identify the intention or purpose of the input information. For example, if the first input information is "My mobile phone has been unable to automatically connect to the Bluetooth device via Bluetooth, and I can only use WiFi connection all the time, which makes me very distressed", its intent information is "Bluetooth cannot automatically connect", and the intent recognition method can be obtained by identifying the input information through the intent recognition algorithm in the existing natural language understanding.

[0074] Based on the intent information, select relevant characters, words, sentences and other partial information from the first input information as the first attribute information. When selecting, you can traverse and determine whether the characters, words and sentences in the first input information are associated with the intent information. The associated judgment method can be to input the characters, words, sentences and intent information into the existing knowledge graph to determine whether they belong to the same field. If they belong to the same field, it is determined that the characters, words and sentences are associated with the intent information. Taking the first input information "My mobile phone's WiFi has a problem" as an example, the first attribute information selected from it may be "mobile phone" or "WiFi".

[0075] Determine whether the first attribute information meets the preset condition, wherein the preset condition is a pre-set rule, such as determining whether the amount of the selected first attribute information is sufficient. If it is determined that the first attribute does not meet the preset condition, the server may not be able to accurately reply to the target object's desired result based on the currently selected first attribute information, then the server selects the corresponding reply statement from the pre-stored reply template as the first output information based on the first attribute information, wherein the first output information is preferably information in the form of a rhetorical question, and continues to take the first input information as "My mobile phone WiFi has a problem" as an example, the server queries the first output information from the pre-stored reply template based on the first attribute information - "mobile phone", "WiFi", such as "What's the problem with mobile phone WiFi?", and sends the first output information to the target object, and instructs the target object to feedback the second input information based on the first output information, and the instruction method can be to add fixed words such as "Please answer" on the basis of the first output information to prompt the user.

[0076] In the second output module 203, after receiving the second input information of the target object, the server side performs intent recognition on the second input information to obtain corresponding intent information, and then selects relevant characters, words or sentences and other partial information from the second input information as second attribute information based on the intent information of the second input information. The method for recognizing the intent information of the second input information and the method for selecting the second attribute information are the same as the processing method for the first input information, and will not be elaborated here.

[0077] Determine whether the first attribute information and the second attribute information meet the preset conditions, for example, determine whether the number of the selected first attribute information and the second attribute information is sufficient. If it is determined that the first attribute information and the second attribute information meet the above preset conditions, the first attribute information and the second attribute information can be integrated, wherein the integration method can be the mutual superposition of the attribute information, or the replacement of the attribute information, and determine whether the integrated attribute information still has corresponding intention information. The method for determining whether there is corresponding intention information can be to establish a mapping relationship between multiple attribute information and intention information in advance and store them in the information library. When the integrated attribute information is queried in the information library, it is determined that the integrated attribute information has corresponding intention information. After the integrated attribute information is queried, the corresponding intention information is extracted from the information library, and the reply information is generated according to the extracted intention information, and is used as at least a part of the second output information, wherein the specific method for generating the reply information can be to input the extracted intention information into the dialogue system, and generate the reply information through the dialogue system. Of course, in addition to the reply information, the second output information can also include fixed speech information.

[0078] If the server determines that the first attribute information and the second attribute information still do not meet the preset conditions, the third output information is generated based on the first attribute information and the second attribute information and fed back to the target object, and the above steps are repeated, wherein the generation process of the third output information is the same as the above-mentioned method of generating the first output information, and will not be repeated here. Furthermore, during the feedback process, the server can also set the number of feedbacks on the server side, and the number of feedbacks is accumulated every time the attribute information is determined to not meet the preset conditions. When the accumulated number exceeds the number of feedbacks on the server side, the server side can end the current dialogue and switch to manual service.

[0079] Therefore, the attribute information of the target object scattered in multiple rounds of conversations can be extracted multiple times through information interaction, and the intention can be understood based on all the attribute information extracted in multiple rounds of conversations, thereby improving the accuracy of intelligent customer service feedback information and improving user experience.

[0080] In one possible implementation, the first output module 202 is specifically configured to:

[0081] Extracting first attribute information based on a first preset information set, wherein the first preset information set includes an attribute type corresponding to the attribute information;

[0082] Based on the intention information in the first input information, corresponding first attribute information is extracted from the first input information, wherein the extracted first attribute information corresponds to an attribute type in the first preset information set.

[0083] In this embodiment, after extracting the first attribute information, the server side stores the extracted first attribute information in a first preset information set, wherein the first preset information set is a pre-set data set and a mapping relationship is pre-established with the intent information of the first input information. After identifying the intent information in the first input information, the corresponding first preset information set can be obtained according to the mapping relationship.

[0084] The first preset information set includes at least one attribute type, such as subject, predicate and object. According to the attribute type in the first preset information set, the first attribute information is extracted from the first input information. Taking the preset information set including three attribute types of subject, predicate and object, and the input information being "the mobile phone cannot automatically connect to WiFi" as an example, when extracting the attribute information, the attribute information extracted from the input information is "mobile phone", "cannot automatically connect" and "WiFi". The extraction method can specifically be to obtain the type of each word in the first input information during the process of understanding the intent of the first input information through the existing intent understanding module, match the type of each word with the attribute type, extract the successfully matched words and use them as the first attribute information. The specific extraction method can also be obtained through sequence annotation or multi-label classification, that is, a large number of words are established in advance, and each word is sequence-annotated or labeled, and the words in the first input information are matched with the words after sequence annotation or multi-label classification. If the match is successful, the same word type is set for the words in the first input information, and then the words that have been set to the same word type are matched with the attribute type of the first preset information set, and the successfully matched words are extracted and used as the first attribute information.

[0085] In one possible implementation, the second output module 203 is specifically configured to:

[0086] determining an intent relevance between the first input information and the second input information;

[0087] Determining a corresponding intermediate preset information set based on the intention information of the second input information;

[0088] If it is determined that the intention relevance does not exceed the preset threshold, generating a second preset information set based on the first preset information set and the intermediate preset information set;

[0089] If the determination intention relevance exceeds a preset threshold, the intermediate preset information set is used as the second preset information set;

[0090] Generate second output information according to the second preset information set.

[0091] In this embodiment, the intention relevance is the similarity between the intention information representing the first input information and the intention information of the second input information. The similarity calculation method can be obtained by using the first input information and the second input information as the input of the existing text similarity algorithm in natural language processing for discrimination.

[0092] Meanwhile, based on the intention information of the second input information, a corresponding intermediate preset information set is determined. The intermediate preset information set, like the first preset information set, is a preset data set, and is also pre-established with a mapping relationship with the intention information of the second input information, and also includes at least one attribute type. The specific process of determining the intermediate preset information set is as follows: the corresponding intermediate preset information set is queried through the mapping relationship of the intention information of the second input information, and the second attribute information is extracted from the second input information according to the attribute type in the intermediate preset information set, where the extraction method is the same as the method of extracting the first attribute information above, and will not be repeated here. The extracted second attribute information is stored in the intermediate preset information set according to the attribute type.

[0093] Judge whether the intention relevance exceeds a preset threshold, where the preset threshold is a fixed value and can be adjusted according to actual needs.

[0094] If it is determined that the intention relevance exceeds the preset threshold, it means that the intention relevance between the first input information and the second input information is relatively high, that is, the first input information and the second input information both represent the same thing. Then, the first preset information set and the intermediate preset information set are processed by addition or replacement according to the attribute type. The addition process is as follows: if there is second attribute information of a certain attribute type in the intermediate preset information set, but there is no attribute information of the same attribute type in the first preset information set, then the second attribute information is added to the first preset information set. For example, the first preset information set is <mobile phone, None, WiFi>, and the intermediate preset information set is <None, cannot be automatically connected, None>. After addition, the finally obtained second preset information set is <mobile phone, cannot be automatically connected, WiFi>. The replacement process is as follows: if there is second attribute information of a certain attribute type in the intermediate preset information set, and there is also attribute information of the same attribute type in the first preset information set, then the second attribute information replaces the attribute information in the first preset information set. For example, the first preset information set is <mobile phone, cannot be automatically connected, WiFi>, and the intermediate preset information set is <None, None, Bluetooth>. After replacement, the finally obtained second preset information set is <mobile phone, cannot be automatically connected, Bluetooth>.

[0095] If it is determined that the intention relevance does not exceed the preset threshold, it means that the intention relevance between the first input information and the second input information is low, that is, the first input information and the second input information are two completely unrelated things, then the original first preset information set is discarded, and the intermediate preset information set is directly used as the second preset information set.

[0096] Generate second output information based on the second preset information set; specifically, determine whether there is corresponding intent information in the information library for the attribute information in the current second preset information set. If it is determined that there is corresponding intent information in the information library, generate reply information based on the intent information and use it as at least a part of the second output information. Of course, in addition to the reply information, the second output information may also include fixed speech information.

[0097] Another aspect of the present invention provides an information interaction device, the device comprising a memory and a processor,

[0098] The memory is used to store instructions, and the instructions are used to control the processor to operate to implement the above-mentioned information interaction method during execution.

[0099] In an embodiment of the present invention, when the storage instruction is executed, it is used to receive first input information of a target object; if the first attribute information corresponding to the first input information does not meet the preset condition, first output information is generated according to the first attribute information, and the target object is instructed to feedback second input information according to the first output information; wherein the first attribute information is at least part of the first input information and corresponds to the intention information of the first input information; if the second attribute information corresponding to the first attribute information and the second input information meets the preset condition, second output information is generated according to the first attribute information and the second attribute information and fed back to the target object; wherein the second attribute information is at least part of the second input information and corresponds to the intention information of the second input information, and the second output information includes at least information generated according to the first attribute information and the second attribute information.

[0100] Therefore, the attribute information of the target object scattered in multiple rounds of conversations can be extracted multiple times through information interaction, and the intention can be understood based on all the attribute information extracted in multiple rounds of conversations, thereby improving the accuracy of intelligent customer service feedback information and improving user experience.

[0101] Another aspect of the present invention provides a computer-readable storage medium, the storage medium comprising a group of computer-executable instructions, which are used to execute any of the above-mentioned information interaction methods when the instructions are executed.

[0102] In an embodiment of the present invention, a computer-readable storage medium includes a set of computer-executable instructions, which, when executed, are used to receive first input information of a target object; if the first attribute information corresponding to the first input information does not meet a preset condition, first output information is generated based on the first attribute information, and the target object is instructed to feedback second input information based on the first output information; wherein the first attribute information is at least part of the first input information and corresponds to the intention information of the first input information; if the second attribute information corresponding to the first attribute information and the second input information meets the preset condition, second output information is generated based on the first attribute information and the second attribute information and fed back to the target object; wherein the second attribute information is at least part of the second input information and corresponds to the intention information of the second input information, and the second output information includes at least information generated based on the first attribute information and the second attribute information.

[0103] Therefore, the attribute information of the target object scattered in multiple rounds of conversations can be extracted multiple times through information interaction, and the intention can be understood based on all the attribute information extracted in multiple rounds of conversations, thereby improving the accuracy of intelligent customer service feedback information and improving user experience.

[0104] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art may combine and combine different embodiments or examples described in this specification and the features of different embodiments or examples, unless they are contradictory.

[0105] In addition, the terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined as "first" and "second" may explicitly or implicitly include at least one of the features. In the description of the present invention, the meaning of "plurality" is two or more, unless otherwise clearly and specifically defined.

[0106] The above are only specific embodiments of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art who is familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed by the present invention, which should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention should be based on the protection scope of the claims.

Claims

1. An information interaction method, the method comprising: Receiving first input information of a target object; If the first attribute information corresponding to the first input information does not meet the preset condition, generating first output information according to the first attribute information, and instructing the target object to feedback second input information according to the first output information; The first attribute information is at least part of the first input information and corresponds to the intention information of the first input information; If the first attribute information and the second attribute information corresponding to the second input information meet the preset condition, generating second output information according to the first attribute information and the second attribute information and feeding back the second output information to the target object; The second attribute information is at least part of the second input information and corresponds to the intention information of the second input information, and the second output information includes at least information generated according to the first attribute information and the second attribute information; Extracting the first attribute information based on a first preset information set, wherein the first preset information set includes an attribute type corresponding to the attribute information; Based on the intention information in the first input information, extracting corresponding first attribute information from the first input information, wherein the extracted first attribute information corresponds to an attribute type in the first preset information set; The generating second output information according to the first attribute information and the second attribute information comprises: determining an intention correlation between the first input information and the second input information; Determining a corresponding intermediate preset information set based on the intention information of the second input information; If it is determined that the intention relevance exceeds a preset threshold, generating a second preset information set according to the first preset information set and the intermediate preset information set; If it is determined that the intention relevance does not exceed the preset threshold, using the intermediate preset information set as the second preset information set; The second output information is generated according to the second preset information set.

2. According to the method of claim 1, the first attribute information corresponding to the first input information does not meet a preset condition, including at least one of the following: The amount of the first attribute information is lower than a first threshold; The first attribute information does not correspond to an attribute type in the first preset information set.

3. According to the method of claim 1, the intermediate preset information set includes multiple attribute types; Correspondingly, generating the second preset information set according to the first preset information set and the intermediate preset information set includes: storing the second attribute information into the intermediate preset information set according to the attribute type; According to the second attribute information in the intermediate preset information set, the first attribute information in the first preset information set is replaced, or the second attribute information is added to the first preset information set.

4. The method according to claim 1, wherein generating second output information according to the first attribute information and the second attribute information comprises: Based on the first attribute information and the second attribute information, query corresponding intent information from an information database; Generate second output information based on the queried intention information.

5. According to the method of claim 1, the method of obtaining the first attribute information / the second attribute information comprises: Acquired by understanding the intention of the first input information / the second input information; Obtained through sequence labeling or multi-label classification.

6. An information interaction device, comprising: An information receiving module, used for receiving first input information of a target object; a first output module, configured to generate first output information according to the first attribute information if the first attribute information corresponding to the first input information does not meet a preset condition, and instruct the target object to feed back second input information according to the first output information; The first attribute information is at least part of the first input information and corresponds to the intention information of the first input information; A second output module, configured to generate second output information according to the first attribute information and the second attribute information and feed it back to the target object if the second attribute information corresponding to the first attribute information and the second input information meets the preset condition; The second attribute information is at least part of the second input information and corresponds to the intention information of the second input information, and the second output information includes at least information generated according to the first attribute information and the second attribute information; The first output module is further used to extract the first attribute information based on a first preset information set, wherein the first preset information set includes an attribute type corresponding to the attribute information; Based on the intention information in the first input information, extracting corresponding first attribute information from the first input information, wherein the extracted first attribute information corresponds to an attribute type in the first preset information set; The second output module is further used to determine the intention correlation between the first input information and the second input information; Determining a corresponding intermediate preset information set based on the intention information of the second input information; If it is determined that the intention relevance exceeds a preset threshold, generating a second preset information set according to the first preset information set and the intermediate preset information set; If it is determined that the intention relevance does not exceed the preset threshold, using the intermediate preset information set as the second preset information set; The second output information is generated according to the second preset information set.

7. An information interaction device, comprising a memory and a processor, The memory is used to store instructions, and the instructions are used to control the processor to operate so as to implement the information interaction method described in any one of claims 1-5 when executed.

8. A readable storage medium, comprising a set of computer executable instructions, which are used to execute the information interaction method according to any one of claims 1 to 5 when the instructions are executed.

Images