Question and answer information processing method, commodity information display method, equipment and storage medium

Abstract

Embodiments of the present application provide a question and answer information processing method, a commodity information display method, equipment and a storage medium. In the question and answer information processing method, in a human-to-human consultation dialogue scene, at least one round of consultation dialogue between people for a target object can be used for question and answer recognition, so as to obtain object information of the target object. The object information can be displayed in the detail information of the target object after being confirmed by a responding user. Based on this way, the depth of the dialogue information between people is mined, effective information can be mined from the dialogue information and reasonably used, which is beneficial to reduce the time cost and labor cost required for repeated consultation and repeated consultation reply.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of computers, and particularly relates to a question and answer information processing method, a commodity information display method, equipment and a storage medium. BACKGROUND

[0002] With the development of electronic commerce technology, online shopping of consumers and online transactions between merchants are becoming more and more widespread by means of information network technology. In the online transaction scenario, sellers often have a conversation chat related to commodities, especially in the second-hand trading platform. The chat conversation between the buyer and the seller contains rich commodity information, but the reuse rate of the commodity information is low. SUMMARY

[0003] Aspects of the present application provide a question and answer information processing method, a commodity information display method, equipment and a storage medium to improve the reuse rate of effective information generated in the conversation chat.

[0004] The present application provides a question and answer information processing method, comprising: receiving object information of a target object sent by a server; the object information is obtained according to at least one round of consultation conversation between a consultation user and a response user; in a consultation conversation interface of the target object, the object information is displayed to the response user; in response to a confirmation operation of the response user on the object information, an addition request is sent to the server, so that the server adds the object information to the detail information of the target object.

[0005] The present application also provides a commodity information display method, comprising: obtaining consultation conversation information corresponding to a commodity; performing question and answer information identification based on the consultation conversation information to obtain commodity description information of the commodity; and taking the commodity description information as display information of the commodity to display the commodity.

[0006] The present application also provides a question and answer information processing method, comprising: obtaining at least one round of consultation conversation between a consultation user and a response user for a target object; performing question and answer information identification on the at least one round of consultation conversation to obtain object information of the target object; sending the identified object information to a terminal device of the response user, so that the terminal device displays the object information to the response user; and receiving an addition request returned by the terminal device according to a confirmation operation of the response user on the object information, and adding the object information to the detail information of the target object.

[0007] The embodiment of the present application further provides a terminal device, comprising a memory, a processor and a communication component; the memory is used for storing one or more computer instructions; the processor is used for executing the one or more computer instructions to perform the steps in the method provided by the embodiment of the present application.

[0008] The embodiment of the present application further provides a server, comprising a memory, a processor and a communication component;

[0009] The memory is used for storing one or more computer instructions; the processor is used for executing the one or more computer instructions to perform the steps in the method provided by the embodiment of the present application.

[0010] The embodiment of the present application further provides a computer readable storage medium storing a computer program, the computer program is executed to implement the steps in the method provided by the embodiment of the present application.

[0011] In the embodiment of the present application, in the human-to-human consultation dialogue scenario, the question and answer recognition can be performed according to at least one round of consultation dialogue between people for a target object, so as to obtain object information of the target object. The object information can be displayed in the detail information of the target object after being confirmed by the answering user. Based on this way, the depth mining of the dialogue information between people is realized, the effective information can be mined from the dialogue information and reasonably utilized, and the time cost and the labor cost required for repeated consultation and repeated consultation reply are reduced. BRIEF DESCRIPTION OF DRAWINGS

[0012] The accompanying drawings for explaining the present application are used to provide further understanding of the present application, constitute a part of the present application, the illustrative embodiments of the present application and the explanation thereof are used to explain the present application, and do not constitute an improper limitation on the present application. In the drawings:

[0013] Figure 1 A structural schematic diagram of a question and answer information processing system provided for an exemplary embodiment of the present application;

[0014] Figure 2 A processing flowchart of a question and answer information recognition model provided for an exemplary embodiment of the present application;

[0015] Figure 3 A schematic diagram of a consultation dialogue interface of a target object provided for an exemplary embodiment of the present application;

[0016] Figure 4 A schematic diagram of a structured attribute information display structure in a consultation dialogue interface of a target object provided for an exemplary embodiment of the present application;

[0017] Figure 5A schematic diagram showing a question-and-answer pair in a consultation dialogue interface of a target object, provided as another exemplary embodiment of this application;

[0018] Figure 6 A schematic diagram of a question-and-answer recognition process in a commodity transaction scenario provided as an exemplary embodiment of this application;

[0019] Figure 7 A flowchart illustrating a question-and-answer information processing method provided in an exemplary embodiment of this application;

[0020] Figure 8 A flowchart illustrating a question-and-answer information processing method provided in another exemplary embodiment of this application;

[0021] Figure 9 A schematic diagram of the structure of a terminal device is provided as another exemplary embodiment of this application;

[0022] Figure 10 This is a schematic diagram of the structure of a server provided as another exemplary embodiment of this application. Detailed Implementation

[0023] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions of this application will be clearly and completely described below in conjunction with specific embodiments and corresponding drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0024] In online transactions, sellers typically engage in conversations related to the product, especially on secondhand trading platforms. These conversations contain a wealth of product information; however, the reuse rate of this information is low.

[0025] To address this technical problem, this application provides a question-and-answer information processing method in some exemplary embodiments, specifically for "human-to-human dialogue" interaction scenarios. This method can extract effective information during the dialogue process, thereby achieving the rational utilization of dialogue information. The optional implementation methods of various embodiments of this application will be described in detail below with reference to the accompanying drawings.

[0026] Figure 1 This is a schematic diagram of the structure of a question-and-answer information processing system provided as an exemplary embodiment of this application. Figure 1 As shown, the question-and-answer information processing system 10 includes: a first terminal device 11, a second terminal device 12, and a server 13.

[0027] In this embodiment, the terminal device refers to a device capable of interacting with a user. In some scenarios, the terminal device can be implemented as a smartphone, a tablet computer, a computer device, a smart wearable device, etc.

[0028] In this embodiment, the first terminal device 11 and the second terminal device 12 are different terminal devices and can be held by different users. In this embodiment, for ease of description, the user holding the first terminal device 11 is described as a consulting user, and the user holding the second terminal device 12 is described as a responding user. The consulting user and the responding user can be implemented as different dialogue roles in a "person-to-person dialogue" scenario. For example, in an online shopping scenario, the consulting user can be implemented as a buyer user, and the responding user can be implemented as a seller user (including a seller customer service, a seller himself, etc.). In an online consultation scenario, the consulting user can be implemented as a consulting visitor, and the responding user can be implemented as a consultant (such as a psychological consultant, a doctor, a lawyer), etc.

[0029] In this embodiment, the first terminal device 11 and the second terminal device 12 are used to interact with the respective users and communicate with the server 13. The server 13 can forward data between the first terminal device 11 and the second terminal device 12 and process the question-and-answer information based on the dialogue data sent by the first terminal device 11 and the second terminal device 12. This embodiment does not limit the implementation form of the server 13, which can be a conventional server, a cloud server, a cloud host, a virtual center, etc.

[0030] In this embodiment, the server device mainly includes a processor, a hard disk, a memory, a system bus, etc., which are similar to the general computer architecture. In addition, the server 13 and the first terminal device 11 and the second terminal device 12 can be connected by wireless or wired connection. In this embodiment, if the first terminal device 11 and the second terminal device 12 are connected to the server 13 through a mobile network, the network standard of the mobile network can be any one of 2G (GSM), 2.5G (GPRS), 3G (WCDMA, TD-SCDMA, CDMA2000, UTMS), 4G (LTE), 4G+ (LTE+), 5G, WiMax, etc., which is not limited in this embodiment.

[0031] The following part will introduce the optional implementation mode of the server 13 for processing the question-and-answer information in detail.

[0032] In the question and answer information processing system 10, when the consulting user needs to consult the target object, the consulting user can initiate a conversation through the first terminal device 11, and the answering user can respond to the conversation of the consulting user based on the second terminal device 12. Generally, the consulting conversation includes one conversation round, or the consulting conversation can last for multiple conversation rounds. One conversation round can include one speech of the consulting user and one speech of the answering user. The target object being consulted can be an entity object (such as a commodity) or a virtual object (such as a medical service), and the embodiment includes but is not limited to this.

[0033] The server 13 can obtain at least one round of consulting conversation of the consulting user and the answering user for the target object, and perform question and answer information identification on the at least one round of consulting conversation to obtain object information of the target object.

[0034] When the server 13 performs question and answer information identification, the server 13 can extract a sentence for asking a property of the target object and a sentence for answering the property of the target object from the at least one round of conversation based on an MTQAE (Multi-Turn Question and Answer Extraction) algorithm model. The specific implementation process of the MTQAE algorithm model will be introduced in subsequent embodiments, and will not be described here.

[0035] In some embodiments, the server 13 can obtain at least one round of consulting conversation between the consulting user and the answering user after the consulting user and the answering user end the conversation, and start question and answer information identification. The at least one round of conversation can be part of the conversation or all of the conversation generated in the entire consulting conversation process.

[0036] In another embodiment, the server 13 can obtain at least one round of consulting conversation between the consulting user and the answering user according to a period, and start question and answer information identification. The at least one round of conversation can be all of the conversation generated in the conversation period. The length of the period can be set according to requirements, for example, 15 seconds, 30 seconds, or 1 minute, and the embodiment is not limited.

[0037] In yet another embodiment, at least one round of consulting conversation between the consulting user and the answering user can be obtained and question and answer information identification can be started each time a reply message of the answering user is obtained. The at least one round of conversation can be the conversation of a specified round before the reply message (including the reply message).

[0038] The object information recognized by the server 13 can be represented in a dialogue form or in a summarized information form, and the embodiments are not limited in this regard. For example, in a commodity consultation scenario, if the object information recognized by the server 13 is represented in a dialogue form, the object information can be: “Question: What is the material of this clothes? Answer: It is pure cotton.” If the object information recognized by the server 13 is represented in a summarized information form, the object information can be: “Material: Pure cotton.”

[0039] After the object information of the target object is recognized based on the at least one round of consultation dialogue, the server 13 can send the recognized object information to the second terminal device 12, so as to display the object information to the answering user on the second terminal device 12.

[0040] After the second terminal device 12 receives the object information of the target object sent by the server 13, the second terminal device 12 can display the object information to the answering user in a consultation dialogue interface of the target object. The consultation dialogue interface of the target object is an interface provided by the second terminal device 12 for communicating with the consulting user.

[0041] In response to a confirmation operation of the answering user on the object information, the second terminal device 12 sends an adding request to the server 13. The confirmation operation of the answering user can be triggered by a physical button on the second terminal device 12 (for example, a confirmation button of a mobile phone or an enter button of a computer), a voice instruction, or a click operation on a control (for example, a confirmation control) displayed on the second terminal device 12, and the embodiments include but are not limited to the above.

[0042] After receiving the adding request, the server 13 can add the object information to the detail information of the target object. Thus, when the consulting user, the answering user, or other users request to view the detail information of the target object from the server 13, the object information recognized based on the at least one round of consultation dialogue can be viewed in the detail information.

[0043] In the embodiments, in a person-to-person consultation dialogue scenario, the object information of the target object can be obtained through question and answer recognition based on at least one round of consultation dialogue between the person and the person. The object information can be displayed in the detail information of the target object after being confirmed by the answering user. In this way, the depth of the dialogue information between the person and the person is mined, effective information is mined from the dialogue information and reasonably utilized, and the time cost and the labor cost required for repeated consultation and repeated consultation reply are reduced.

[0044] In the foregoing embodiments, the server 13 initiates the question-answer information recognition in multiple manners: implementation 1: after the consultation user and the response user end the dialogue, the server 13 initiates the question-answer information recognition; implementation 2: the server 13 initiates the question-answer information recognition periodically; and implementation 3: after receiving the reply message of the response user, the server 13 initiates the question-answer information recognition.

[0045] In the implementation 3, if the number of speeches of the response user is large, the server 13 needs to initiate the question-answer information recognition operation multiple times. To reduce the calculation amount and frequency of the question-answer information recognition, the server 13 can further detect the interactive intention of the consultation user, and can perform the implementation 3 when it is detected that the interactive intention of the consultation user is used to inquire the attribute information of the target object. The following will be further exemplarily described.

[0046] Optionally, when the server 13 obtains at least one round of consultation dialogue between the consultation user and the response user, the server 13 can recognize the interactive intention of the consultation user according to the user question of the consultation user. When the server 13 recognizes the interactive intention, the server 13 can use a binary classification algorithm to determine whether the speech (for example, the question message) of the consultation user is used to inquire the related information of the target object. The binary classification algorithm can be a logistic regression algorithm, a fast-text algorithm, and the like, but is not limited thereto.

[0047] If the interactive intention of the consultation user is used to inquire the attribute information of the target object, the server 13 can obtain the dialogue of a specified number of rounds before the reply message of the response user as at least one round of consultation dialogue for question-answer recognition each time the reply message of the response user is received. The at least one round of consultation dialogue includes the reply message.

[0048] For example, in a shopping consultation scenario, after the server 13 obtains the speech of the consultation user (that is, the buyer), the server 13 can use a binary classification algorithm to determine whether the speech of the buyer is used to inquire the information related to the goods. If the information related to the goods is inquired, the server 13 can obtain the dialogue of N rounds before the reply message of the response user (that is, the seller) as the dialogue for question-answer recognition after the reply message of the response user is obtained. N is a positive integer, and the value of N can be set according to actual needs. In some embodiments, N can be set to 8, 10, 12, and the like, but the present embodiment is not limited thereto.

[0049] It should be noted that in the foregoing and the following embodiments of the present application, the at least one round of consultation dialogue obtained by the server 13 can include complete dialogue information between the consultation user and the response user, and the dialogue information includes the speech round, the speech time, the speech role, and the speech content of the consultation user and the response user, and the like. The foregoing dialogue information can cover the context information of the dialogue, thereby facilitating the mining of the effective information hidden in the dialogue.

[0050] After obtaining the at least one round of consultation dialogue between the consultation user and the answering user, the server 13 can perform question and answer information recognition on the at least one round of consultation dialogue to obtain at least one object information of the target object. The question and answer information recognition is implemented based on an MTQAE algorithm model. The following will be exemplarily described.

[0051] Optionally, the server 13 can input the at least one round of consultation dialogue into a question and answer extraction model. In the question and answer extraction model, tokenization is performed on a plurality of sentences contained in the at least one round of consultation dialogue to obtain tokenization results of the plurality of sentences respectively. The tokenization refers to splitting a sentence according to a word / vocabulary library, word / root, word / suffix, etc. defined by an NLP (natural language processing) model to obtain a string of tokens corresponding to the sentence.

[0052] After obtaining the tokenization results of the plurality of sentences respectively, the question and answer extraction model can perform semantic encoding on the tokenization results of the plurality of sentences respectively to obtain encoding vectors of the plurality of sentences respectively. In some optional embodiments, the semantic encoding can be implemented based on a Bert (a natural language processing model) or a Transformer (a natural language processing model framework) family model, or can be implemented based on an RNN (Recurrent Neural Network, recurrent neural network) or an LSTM (Long Short-Term Memory, long short-term memory neural network), a Word2Vec model, etc. The embodiments contain but are not limited to the above.

[0053] Optionally, when performing semantic encoding on the tokenization results of each sentence, at least one of the speaking role and the keyword of the sentence can be further combined, so as to strengthen the semantic information.

[0054] After obtaining the encoding vectors of the plurality of sentences respectively, the question and answer extraction model can perform fusion processing on the encoding vectors of the plurality of sentences respectively by using at least one of the speaking role, the question and answer position, and the speaking time of the plurality of sentences to obtain fusion vectors of the plurality of sentences respectively.

[0055] The following will exemplarily describe a method for obtaining the fusion vector of the sentence by taking any one of the plurality of sentences as an example.

[0056] Optionally, for any of the plurality of sentences, in the question-answer extraction model, the encoding vector of the sentence, the speaking role corresponding to the sentence, and the keyword information of the sentence can be spliced to obtain a first feature vector of the sentence. In some embodiments, the keyword information of the sentence can include keyword hit information of the sentence, which is used to describe whether the sentence hits the set keyword.

[0057] It should be noted that, in some embodiments, before splicing the encoding vector of the sentence, the speaking role corresponding to the sentence, and the keyword of the sentence, the Embedding (a coding method for converting discrete variables into continuous vectors) technology can be used to encode the speaking role and the keyword hit information of the sentence to obtain an Embedding vector corresponding to the speaking role and an Embedding vector of the keyword hit information of the sentence. When splicing, the encoding vector of the sentence, the Embedding vector corresponding to the speaking role, and the Embedding vector of the keyword hit information can be spliced to obtain the first feature vector.

[0058] After obtaining the first feature vector, the first feature vector is positionally encoded with the question-answer position information of the sentence in the at least one round of question and answer to obtain a second feature vector. In some optional embodiments, the algorithm of the positional encoding can be implemented based on the algorithm of the relative position encoding, or can be implemented based on the absolute position encoding, the rotational position encoding, etc. The present embodiment does not make any limitation. Optionally, the operation of the positional encoding can be implemented based on the position (round) information fusion layer in the MTQAE model.

[0059] After obtaining the second feature vector, the question-answer extraction model can input the second feature vector into a multi-head attention network to obtain a third feature vector of the sentence. In the multi-head attention network, deep information interaction and encoding can be performed to mine deeper text information. After obtaining the third feature vector, the question-answer extraction model splices the third feature vector with the speaking time information of the sentence to obtain a fourth feature vector as the fusion vector of the sentence. Optionally, the speaking time information of the sentence can include at least one of the following: a round difference between the sentence and the nearest sentence on the opposite side, a time difference between the sentence and the nearest sentence on the opposite side, a time difference between the sentence and the nearest sentence on the same side, and a round difference between the sentence and the nearest sentence on the same side. In the present embodiment, the speaking time is fused into the feature, which can capture the influence of the dialogue timing on the communication content according to the dialogue characteristics between people, and improve the recognition accuracy.

[0060] Optionally, before splicing the third feature vector and the speaking time information of the sentence, the speaking time information can be discretized and encoded by using the Embedding technology to obtain an Embedding vector corresponding to the speaking time information. When splicing the third feature vector and the speaking time information of the sentence, the third feature vector and the Embedding vector corresponding to the speaking time information can be spliced.

[0061] The question and answer extraction model can perform the above implementation for each of the plurality of sentences to obtain a fusion vector of each of the plurality of sentences. After obtaining the fusion vector of each of the plurality of sentences, the question and answer extraction model can identify a question sentence and a reply sentence related to the attribute of the target object from the plurality of sentences based on the fusion vector of each of the plurality of sentences, and obtain at least one object information of the target object from the question sentence and the reply sentence related to the attribute of the target object.

[0062] The operation of identifying the question sentence and the reply sentence related to the attribute of the target object based on the fusion vector of each of the plurality of sentences can be implemented based on a multi-label classifier (Multi-Label Classification). Before inputting the fusion vector into the multi-label classifier, an intermediate layer (Intermediate Layer) can be inputted for dimension reduction.

[0063] The following will be combined with Figure 2 A round of dialogue between a buyer and a seller in a commodity transaction scenario will be taken as an example to further exemplarily illustrate the above calculation process. A round of dialogue between the buyer and the seller contains one question and one answer, and the turn, time, role and content of the two sentences are respectively as follows: Figure 2As shown. After the two conversations are cut and semantically encoded, the encoded vectors V1 and V2 are obtained respectively. Next, V1 can be spliced with the Embedding vector of the speaking role R1 corresponding to the first sentence and the Embedding vector K1 of the keyword hit information to obtain the feature vector V1`; V2 can be spliced with the speaking role R2 corresponding to the second sentence and the Embedding vector K2 of the keyword hit information to obtain the feature vector V2`. Next, based on the position coding module, the speaking turns 0 and 1 are combined to perform position coding on the feature vectors V1` and V2` respectively to obtain the feature vectors V1`` and V2``. Next, the multi-head attention network is used to encode the feature vectors V1`` and V2`` to obtain the feature vectors V1``` and V2```. The feature vectors V1``` and V2``` can be spliced with the Embedding vectors of the speaking times t1 and t2 respectively to obtain the fusion vectors. The fusion vectors can be dimensionally reduced by the intermediate layer and can be input into the multi-label classifier to obtain the classification recognition results of the two conversations. The classification recognition result of any sentence is used to describe the probability of the sentence for inquiring about the commodity and the probability of the sentence for replying to the commodity information. Based on this algorithm, the conversation about the commodity information question and answer can be identified from the consultation conversation. In this implementation, based on the MTQAE algorithm model, the influence of the role information, speaking turn, speaking time and other information in the conversation information on the recognition result can be fully considered, which is beneficial to more accurately mining the effective information contained in the conversation.

[0064] In some optional embodiments, when at least one object information of the target object is obtained from the question sentence and the reply sentence related to the attribute of the target object, the server 13 can perform matching of the question and answer pair and determine the output mode of the object information.

[0065] Optionally, the server 13 can perform matching of the question and answer pair on the question sentence and the reply sentence related to the attribute of the target object to obtain at least one set of question and answer pairs. Before the matching of the question and answer pair is performed, a finite-state machine (FSM) can be generated based on the speaking time, turn and set prior rules of the question sentence and the reply sentence. The set prior rules can include that the question sentence appears before the reply sentence, no other questions appear between multiple replies to the same question, etc. Based on the FSM, the matching question and answer can be screened from the question sentence and the reply sentence, so as to avoid the identification error of “answering the wrong question”.

[0066] For example, the server 13 identifies the question sentences Q1, Q2 and the answer sentences A1, A2, A3 from the 10 rounds of conversations. Among them, the speaking order is: Q1, A1, Q2, A2, A3. For the answer sentence A1, the speaking order and the question sentence Q1 closest to the answer sentence can be matched as the question of the answer sentence. For the question sentence Q2, there are answer sentences A2, A3 after the question sentence Q2, if the speaking time of the answer sentences A2, A3 is close (for example, less than 5 seconds or 10 seconds, etc.), and there is no other question sentence after the answer sentences A2, A3, then the answer sentences A2, A3 can be matched as the answer of the question sentence Q2.

[0067] Based on the above manner, the server 13 can identify the matched question and answer sentence combination from the identified question and answer sentences related to the target object attribute. Each group can be referred to as a group of question and answer pairs. A group of question and answer pairs can include one question sentence and one answer sentence, i.e. "one question and one answer"; can include one question sentence and multiple answer sentences, i.e. "one question and multiple answers"; can include multiple question sentences and one answer sentence, i.e. "multiple questions and one answer"; or can include multiple question sentences and multiple answer sentences, i.e. "multiple questions and multiple answers".

[0068] Further optionally, after the server 13 obtains at least one group of question and answer pairs based on the foregoing embodiments, for any group of question and answer pairs, if the question and answer pair includes multiple questions, the server 13 can assemble or splice the question and answer pair. For example, if the question and answer pair includes multiple questions Q1, Q2 and one answer A1, the server 13 can splice the multiple questions Q1, Q2 to obtain the question and answer pair (Q1+Q2, A1). If the question and answer pair includes one question Q1 and multiple answers A1, A2, A3, the server 13 can splice the multiple answers A1, A2, A3 to obtain the question and answer pair (Q1, A1+A2+A3).

[0069] For example, assuming that a question and answer pair includes the question Q1, Q2 and the answer A1, the server 13 can splice the questions Q1, Q2 to obtain the question and answer pair (Q1+Q2, A1). Assuming that a question and answer pair includes the question Q1 and the answers A1, A2, A3, the server 13 can splice the answers A1, A2, A3 to obtain the question and answer pair (Q1, A1+A2+A3).

[0070] After the server 13 matches at least one group of question and answer pairs, the server 13 can determine the output mode of each group of question and answer pairs according to the semantic information of each group of question and answer pairs. The output mode includes structured output or unstructured output. The structured output refers to outputting the question and answer pair as information that can be summarized in the structure of "attribute-attribute value".

[0071] Optionally, in some embodiments, a plurality of structured attributes can be set for the target object in advance to describe or introduce the target object. For example, in a commodity transaction scenario, a plurality of structured attributes such as brand, color, size, material, fineness, shelf life, etc. can be set for a commodity.

[0072] Optionally, for any one of the at least one group of question and answer pairs, when determining the output mode of the question and answer pair, the question and answer pair can be subjected to semantic recognition to obtain the semantics of the question and answer pair. If there is an attribute in the preset structured attributes of the target object that matches the semantics of the question and answer pair, it is determined that the structured output mode is used to output the question and answer pair.

[0073] For example, in a commodity transaction scenario, the structured attributes preset for a piece of clothing include brand, color, style, and material. Assuming that the semantics of a question and answer pair is "Q: What is the material of the clothes? A: Pure cotton". There is a "material" attribute in the preset structured attributes of the clothes that matches the semantics of the question and answer pair, so the structured output mode can be used to output the question and answer pair.

[0074] On the contrary, if there is no attribute in the preset structured attributes of the target object that matches the semantics of the question and answer pair, it is determined that the unstructured output mode is used to output the question and answer pair.

[0075] For example, in a second-hand shopping scenario, the semantics of a question and answer pair is "Q: How long has this piece of clothing been bought? A: Bought last summer". This question and answer pair cannot be mapped to the existing structured attributes of the clothes, and the unstructured output mode can be used to output the question and answer pair.

[0076] After determining the output modes of the at least one group of question and answer pairs, the server 13 can output the at least one group of question and answer pairs as at least one object information of the target object.

[0077] Optionally, if the structured output mode is used to output the question and answer pair, the server 13 can perform attribute extraction on the question and answer pair to obtain a target attribute and a corresponding target attribute value of the question and answer pair; and form structured information of the target attribute and the target attribute value as the object information corresponding to the question and answer pair.

[0078] In different application scenarios, different attribute extraction algorithms can be used for attribute extraction of question and answer pairs. For example, in a new product shopping scenario, a new product attribute extraction algorithm can be used for attribute extraction of question and answer pairs. In a second-hand shopping scenario, a second-hand attribute extraction algorithm can be used for attribute extraction of question and answer pairs.

[0079] Optionally, for any attribute of the target object, a keyword can be set for the attribute, and keywords of different attribute values corresponding to the attribute can be set. When performing attribute extraction, the server 13 can determine whether the question-answer pair hits the keyword preset for the attribute of the target object. If a question-answer pair hits the keyword of an attribute, the hit attribute can be determined as the target attribute of the question-answer pair.

[0080] Next, the server 13 can determine whether the question-answer pair hits the keyword preset for different attribute values of the target attribute. If the question-answer pair hits the keyword of an attribute value of the target attribute, the hit attribute value can be determined as the target attribute value of the question-answer pair.

[0081] For example, assuming that the question-answer pair is “question: What is the fabric of the clothes? answer: Pure cotton”. The preset attributes of the clothes include: brand, origin, material, size, and how new. Among them, the keywords of the material attribute can include: material, fabric, and the like. When the above question-answer pair is used for attribute matching, the target attribute of “material” can be matched. Among them, the attribute values of the material attribute can include: cotton, hemp, silk, wool, and silk, etc. When the above question-answer pair is used for attribute value matching, the target attribute value of “cotton” can be matched. Thus, the structured output information of the question-answer pair can be determined as: material-cotton.

[0082] Optionally, if the question-answer pair is output in an unstructured output manner, the server 13 can further perform information security processing on the question-answer pair, and obtain the object information corresponding to the question-answer pair through information security processing. The information security processing operation can include: identifying the private information in the question-answer pair based on the NER (Named Entity Recognition, named entity recognition or proper noun recognition) algorithm, and shielding or blurring the private information. The private information can include: mobile phone number, address, identity card information, and other personal information in the question-answer pair, which is not limited in the embodiment.

[0083] It is also worth mentioning that in some exemplary embodiments, after the server 13 identifies the object information based on the consultation dialogue, the server 13 can further integrate the information in combination with the existing object information to avoid outputting repeated information or conflicting information.

[0084] Optionally, after the server 13 forms the structured information of the target attribute and the target attribute value as the object information corresponding to the question-answer pair, the server 13 can query the existing attribute value of the target attribute from the attributes already displayed in the detail page of the target object; if the existing attribute value of the target attribute is different from the target attribute value, the server 13 can return information prompting to modify the attribute value of the target attribute to the second terminal device 12.

[0085] Optionally, after the server 13 obtains the question-and-answer pair obtained through information security processing as the object information corresponding to the question-and-answer pair, it can determine the target historical question that is the same as the question in the question-and-answer pair from the historical questions displayed in the details page of the target object; determine whether the semantics of the answer to the target historical question is the same as the semantics of the answer to the question-and-answer pair; if so, the server 13 can send the answer to the question-and-answer pair as the answer to the target historical question to the second terminal device 12.

[0086] After obtaining the object information of the target object based on the above implementation method, the server 13 can send the object information to the second terminal device 12 for display.

[0087] After receiving the object information of the target object sent by the server 13, the second terminal device 12 can display the object information to the responding user in the consultation dialogue interface of the target object.

[0088] Optionally, when the second terminal device 12 displays information about the target object in the consultation dialogue interface, it may display a prompt message for supplementary information in the consultation dialogue interface. In response to the triggering operation of this prompt message, a floating window is displayed. This floating window can be located at the bottom of the consultation dialogue interface to avoid obscuring the dialogue content.

[0089] like Figure 3 As shown, in a product transaction scenario, when a buyer inquires about a product with a seller, both the buyer's and seller's terminal devices can display a consultation dialogue interface, showing the conversation between the buyer and seller. When the seller's terminal device receives object information from server 13, it can display a supplementary information prompt message in the target object's consultation dialogue interface, such as... Figure 3 The right-hand image illustrates the prompt message, "To further refine the product description, please add this Q&A to the details." Sellers can decide whether to add product information based on their needs. If there is a need to add product information, the seller can trigger this prompt message. If the seller's terminal device detects the seller's triggering action on this prompt message (e.g., a click), it can display something like... Figure 4 as well as Figure 5 The floating window shown.

[0090] If the object information is output in a structured manner, the second terminal device 12 can display the target attributes and target attribute values ​​corresponding to the object information in the floating window. For example... Figure 4 As shown, in a commodity transaction scenario, the seller terminal can use two target attributes to describe object information. The first target attribute is "style", and its target attribute value is style A; the second attribute is "condition", and its target attribute value is "brand new and never worn".

[0091] If the object information is output in an unstructured manner, the second terminal device 12 can display the question-answer pair corresponding to the object information in the floating window. As shown in the following, Figure 5 In a commodity transaction scenario, the seller terminal can describe the object information by using a question-answer pair, which is "Q: Is it a commemorative edition? A: Yes, it is a commemorative edition released last year."

[0092] Further optionally, when the second terminal device 12 displays the object information to the answering user in the floating window, the second terminal device 12 can further display a modification control of the object information, such as the modification control 401 shown in the following, Figure 4 and the modification control 501 shown in the following. Figure 5

[0093] In response to a triggering operation on the modification control, the second terminal device 12 can obtain the modification result of the object information by the answering user as updated object information. For example, as shown in the following, Figure 4 The seller user can trigger the control corresponding to "Style B" to modify the style of the commodity to "Style B". For another example, as shown in the following, Figure 5 The seller user can trigger the modification control 501 to customize the reply sentence, and details are not described herein.

[0094] After obtaining the updated object information, the answering user can trigger the control (for example, the "Confirm and Supplement" control in the following Figure 4 and Figure 5 ) displayed in the floating window for performing a confirmation and supplement operation, so that the second terminal device 12 can send a request of adding the question-answer pair to the detail information of the target object to the server 13 in response to the triggering operation, and details are not described herein.

[0095] The question-answer information processing system provided by the embodiments of the present application can be applied to various human-to-human conversation scenarios, for example, can include but is not limited to a commodity pre-sale consultation scenario, a commodity post-sale consultation scenario, a commodity post-sale evaluation scenario, a live shopping scenario, a medical service consultation scenario, a legal service consultation scenario, a psychological consultation scenario, and the like. Among them, the commodity pre-sale and post-sale consultation scenarios can include pre-sale and post-sale scenarios of new products, or pre-sale and post-sale scenarios of second-hand commodities. Of course, in addition to the above-mentioned scenarios, it can also be applied to other possible scenarios for commodities, which are not listed one by one. In the following, Figure 6 , a typical second-hand commodity pre-sale consultation scenario will be taken as an example for further illustrative description of the embodiments of the present application.

[0096] ​In the second-hand commodity transaction scenario, due to the special second-hand nature of the commodity, the situation of buy-side silent order is less, and most buyers will communicate with the seller user before purchasing the commodity. When the buyer has the intention to purchase, he / she usually initiates interactive chat with the seller through the consultation portal provided by the transaction platform and inquires the detailed information of the commodity from the seller.

[0097] As shown in Figure 6 After the buyer speaks, the server can perform preliminary screening of the buyer's intention through a binary classification algorithm to quickly determine whether the buyer's speech is asking for commodity-related information. Based on this step, the server can filter most of the non-questioning speeches (such as greetings and good wishes) and recall as many question sentences as possible. Thus, the computational load of the subsequent question and answer extraction module can be greatly reduced.

[0098] If the buyer's speech intention is to inquire about the commodity-related information, the server can input the multi-turn conversation to the question and answer extraction model after each seller speech, and perform commodity question and answer information identification operation based on the question and answer extraction model. The multi-turn conversation refers to the seller's speech and the "complete" buyer-seller conversation information before the seller's speech. The "complete" buyer-seller conversation information includes speech turn, speech time, speech role, and speech content, covering the context of the entire conversation.

[0099] The question and answer extraction module is implemented based on the NLU algorithm MTQAE for multi-turn conversation question and answer extraction in human-to-human scenarios. MTQAE can accurately identify question and answer information in real time. As shown in Figure 6 The question and answer extraction model can identify question and answer information related to the commodity itself, address, and real photo, etc.

[0100] After identifying the question and answer sentences in the multi-turn conversation, the server can sort out a complete set of logic to form a finite state machine FSM based on the speech time, turn, and prior rules (such as the requirement that there must be a question before an answer and the answers to the same question cannot have different problems). Then, the server can perform matching of question and answer sentence pairs based on the FSM to avoid recognition errors of "answering the wrong question".

[0101] The matching results of the question and answer sentence pairs may include "one question with multiple answers", "multiple questions with one answer", and "multiple questions with multiple answers". For the above cases, a group or multiple groups of question and answer pairs can be obtained by assembling and splicing multiple questions or multiple answers according to certain algorithms.

[0102] As shown in Figure 6As shown, after obtaining the question-and-answer pairs, the server can further determine whether the questions and answers are structured based on the information disclosure module. This step is mainly used to determine whether the identified question-and-answer information belongs to the category of structured information. During the determination, preliminary semantic recognition can be performed on each question-and-answer pair, and it can be determined whether the semantic recognition result matches the structured attributes of the product. For example, the question-and-answer pair "Q: How new?; A: 95% new" can be mapped to the structured attribute "condition" of the product.

[0103] If the information can be structured, a sentence transformation operation can be performed, that is, converting the question-and-answer pairs into descriptive statements about the product from the seller's first-person perspective. For structured question-and-answer information, a structured attribute extraction algorithm can be used during output to identify and extract the structured attributes contained in the product information. In the second-hand trading scenario, this attribute extraction algorithm can be a second-hand attribute extraction algorithm, which can extract second-hand attributes such as brand, condition, wear and tear, and shelf life. In the new product trading scenario, this attribute extraction algorithm can be a new product attribute extraction algorithm.

[0104] For unstructured question-answer pairs, the server can use the NER algorithm to mask and obscure private information in the question-answer text. This private information includes, but is not limited to, ID card information, mobile phone number information, and unit / building / floor number information in the address.

[0105] Based on the above processing, we can obtain attributes and attribute values ​​used to describe product information, or question-and-answer pairs used to describe product information.

[0106] If this is the seller's first interaction, the extracted structured attribute information can be directly sent to the seller's device. The seller's device can then display a prompt based on the received information, and after the seller's confirmation, the attribute and attribute value can be added to the product details for display. Similarly, privacy-processed question-and-answer pairs can be sent to the seller's device, which will then display a prompt to the seller, and after the seller's confirmation, the question-and-answer pairs can be added to the product details for display.

[0107] If the conversation is information output by the seller in an interaction before, the server can further perform post-processing operations to integrate and cluster the information.

[0108] like Figure 6 As shown, in the post-processing stage, the server can determine whether the structured attributes or question-and-answer pairs identified during the current communication process are similar to those identified during historical communication processes. If the determination result is yes, further information clustering is required to avoid conflicts or duplications. If the determination result is no, information can be directly displayed as described above.

[0109] If the identified structured attribute and attribute value are the same as the attribute and attribute value that have been displayed in the detail page of the commodity, and the attribute value of the structured attribute is the same as the attribute value that have been displayed in the detail page, the seller user is not prompted to supplement information. If the attribute value of the structured attribute is different from the attribute value that have been displayed in the detail page, the seller is prompted whether to modify the displayed attribute value.

[0110] If the semantic of the question is the same as or similar to the semantic of the question that have been displayed in the detail page of the commodity, the identified answer is supplemented to the question that have been displayed in the detail page, and the seller is prompted.

[0111] Before the structured attribute or the question and answer pair is output, the terminal device of the seller can prompt the seller to perform a confirmation operation, so as to ensure the right to know and the right to choose of the user, and prevent useless information or commodity information that is contrary to the subjective will of the seller from being displayed.

[0112] Through the embodiment, in the second-hand transaction scenario, the text content of the question and answer about the commodity information in the chat of the buyer and the seller can be fully mined, and is supplemented to the detail page of the commodity in the form of a structured attribute or a question and answer pair, so as to increase the content richness of the commodity information, and make other buyers better understand the commodity itself, and reduce the trouble caused to the seller by repeated inquiries of different buyers about the same / similar questions about the same commodity.

[0113] The above embodiments describe the system architecture and system functions of the question and answer information processing system provided by the application, and the following part will specifically describe the question and answer information processing method provided by the embodiments of the application with reference to the accompanying drawings.

[0114] Figure 7 is a method flowchart of a question and answer information processing method provided by an exemplary embodiment of the application, and the embodiment can be implemented based on the question and answer information processing system shown in Figure 1 The embodiment mainly describes from the perspective of the terminal device. As shown in Figure 7 The method comprises the following steps.

[0115] In step 701, object information of a target object sent by a server is received; the object information is obtained according to at least one round of consultation dialogue between a consultation user and a response user.

[0116] In step 702, the object information is displayed to the response user in a consultation dialogue interface of the target object.

[0117] Step 703, in response to the confirmation operation of the answering user on the object information, sending an adding request to the server to make the server add the object information into the detail information of the target object.

[0118] In some example embodiments, in the counseling dialogue interface of the target object, one way of displaying the object information to the answering user includes: displaying a prompt message of supplementary information in the counseling dialogue interface of the target object; in response to a triggering operation on the prompt message, displaying a floating window; in the floating window, displaying the target attribute corresponding to the object information and the target attribute value; or, in the floating window, displaying the question and answer pair corresponding to the object information.

[0119] In some example embodiments, in the floating window, displaying the question and answer pair corresponding to the object information further includes: displaying a modification control of the question and answer pair; in response to a triggering operation on the modification control, obtaining the modification result of the answering user on the question and answer pair as an updated question and answer pair.

[0120] In this embodiment, in the counseling dialogue scenario between people, the terminal device can receive the object information recognized by the server according to at least one round of counseling dialogue between people, and display the object information. The object information can be displayed in the detail information of the target object after being confirmed by the answering user. Based on this way, the depth mining of dialogue information between people is realized, effective information can be mined from the dialogue information and reasonably utilized, which is conducive to reducing the time cost and human cost required for repeated counseling and repeated counseling reply.

[0121] Figure 8 is a first round interaction method flowchart of a question and answer information processing method provided by another example embodiment of the present application, which can be implemented based on the question and answer information processing system shown in Figure 1 , mainly from the perspective of the terminal device. As shown in Figure 8 , the method includes:

[0122] Step 801, obtaining at least one round of counseling dialogue between a counseling user and an answering user for a target object.

[0123] Step 802, performing question and answer information recognition on the at least one round of counseling dialogue to obtain object information of the target object.

[0124] Step 803, sending the recognized object information to the terminal device of the answering user to make the terminal device display the object information to the answering user.

[0125] Step 804, receiving an addition request returned by the terminal device according to the confirmation operation of the responding user on the object information, and adding the object information to the detail information of the target object.

[0126] In some example embodiments, a manner of obtaining at least one round of consultation dialogue between the consulting user and the responding user can include: identifying an interaction intention of the consulting user according to a user question of the consulting user; and if the interaction intention is used to inquire attribute information of the target object, obtaining a specified round of dialogue before each time a reply message of the responding user is received as the at least one round of consultation dialogue.

[0127] In some example embodiments, a manner of performing question and answer information identification on the at least one round of consultation dialogue to obtain at least one object information of the target object can include: inputting the at least one round of consultation dialogue into a question and answer extraction model; performing word segmentation processing on a plurality of sentences contained in the at least one round of consultation dialogue respectively in the question and answer extraction model to obtain a word segmentation result of each of the plurality of sentences; performing semantic coding on the word segmentation result of each of the plurality of sentences to obtain an encoding vector of each of the plurality of sentences; performing fusion processing on the encoding vector of each of the plurality of sentences by using at least one of a speaking role, a question and answer position, and a speaking time of each of the plurality of sentences to obtain a fusion vector of each of the plurality of sentences; identifying a question sentence and a reply sentence related to an attribute of the target object from the plurality of sentences according to the fusion vector of each of the plurality of sentences; and obtaining the at least one object information of the target object from the question sentence and the reply sentence related to the attribute of the target object.

[0128] In some example embodiments, a manner of performing fusion processing on the encoding vector of each of the plurality of sentences by using at least one of a speaking role, a question and answer position, and a speaking time of each of the plurality of sentences to obtain a fusion vector of each of the plurality of sentences can include: for any one of the plurality of sentences, concatenating, in the question and answer extraction model, the encoding vector of the sentence, the speaking role corresponding to the sentence, and a keyword of the sentence to obtain a first feature vector of the sentence; performing position coding on the first feature vector and question and answer position information of the sentence in the at least one round of question and answer to obtain a second feature vector; inputting the second feature vector into a multi-head attention network to obtain a third feature vector of the sentence; and concatenating the third feature vector and speaking time information of the sentence to obtain a fourth feature vector as the fusion vector of the sentence.

[0129] In some example embodiments, the manner of obtaining the at least one object information of the target object from the question sentence and the answer sentence related to the attribute of the target object can include: matching the question sentence and the answer sentence related to the attribute of the target object to obtain at least one set of question-answer pairs; determining the output manner of each of the at least one set of question-answer pairs according to the semantic information of each of the at least one set of question-answer pairs; the output manner includes: structured output or unstructured output; and outputting the at least one set of question-answer pairs as the at least one object information of the target object according to the output manner of each of the at least one set of question-answer pairs.

[0130] In some example embodiments, after the matching of the question sentence and the answer sentence related to the target object to obtain at least one set of question-answer pairs, the method further includes: for any one set of question-answer pairs, if the question-answer pair contains multiple questions, splicing the multiple questions in the question-answer pair; and if the question-answer pair contains multiple answers, splicing the multiple answers in the question-answer pair.

[0131] In some example embodiments, the manner of determining the output manner of each of the at least one set of question-answer pairs according to the semantic information of each of the at least one set of question-answer pairs can include: for any one set of question-answer pairs in the at least one set of question-answer pairs, performing semantic recognition on the question-answer pair to obtain the semantics of the question-answer pair; if there is an attribute in the pre-set structured attribute of the target object that matches the semantics of the question-answer pair, determining to output the question-answer pair in the structured output manner; and if there is no attribute in the pre-set structured attribute of the target object that matches the semantics of the question-answer pair, determining to output the question-answer pair in the unstructured output manner.

[0132] In some example embodiments, the manner of outputting the at least one set of question-answer pairs as the at least one object information of the target object according to the output manner of each of the at least one set of question-answer pairs can include: if the question-answer pair is output in the structured output manner, performing attribute extraction on the question-answer pair to obtain the target attribute and the corresponding target attribute value of the question-answer pair; forming the structured information of the target attribute and the target attribute value as the object information corresponding to the question-answer pair; and if the question-answer pair is output in the unstructured output manner, performing information security processing on the question-answer pair, and obtaining the question-answer pair after the information security processing as the object information corresponding to the question-answer pair.

[0133] In some example embodiments, after forming the structured information of the target attribute and the target attribute value as the corresponding object information of the question-answer pair, the method further includes: querying the existing attribute value of the target attribute from the attributes already shown in the detail page of the target object; if the existing attribute value of the target attribute is different from the target attribute value, returning information prompting modification of the attribute value of the target attribute to the terminal device.

[0134] In some example embodiments, after obtaining the question-answer pair subjected to information security processing as the object information corresponding to the question-answer pair, the method further includes: determining a target historical question identical to the question in the question-answer pair from the historical questions already shown in the detail page of the target object; judging whether the semantic meaning of the answer of the target historical question is identical to the semantic meaning of the answer of the question-answer pair; if yes, sending the answer of the question-answer pair as the answer of the target historical question to the terminal device.

[0135] In the present embodiment, in the scenario of a person-to-person consultation dialogue, the question-answer recognition can be performed according to at least one round of consultation dialogue between persons for a target object, so as to obtain object information of the target object. The object information can be displayed in the detail information of the target object after being confirmed by a responding user. Based on this manner, the deep mining of dialogue information between persons is realized, the effective information can be mined from the dialogue information and reasonably utilized, and the time cost and human cost required for repeated consultation and repeated consultation reply are reduced.

[0136] The present embodiment also provides a commodity information display method, which can be executed by a terminal device of a seller. The method mainly includes the following steps:

[0137] S1, obtaining consultation dialogue information corresponding to a commodity.

[0138] S2, performing question-answer information recognition based on the consultation dialogue information, to obtain commodity description information of the commodity.

[0139] S3, displaying the commodity description information as display information of the commodity, to display the commodity.

[0140] The consultation dialogue information of the commodity is generated in a consultation process of the commodity. The consultation of the commodity can occur in various consultation scenarios for the commodity, such as a pre-sale consultation scenario of the commodity, an after-sale consultation scenario of the commodity, an after-sale evaluation scenario of the commodity, a live shopping scenario, and the like. The pre-sale and after-sale consultation scenarios of the commodity can include pre-sale and after-sale scenarios of new products, or pre-sale and after-sale scenarios of second-hand commodities. Of course, in addition to the above-mentioned scenarios, the other possible scenarios for the commodity can also be applicable, which are not listed one by one. When the buyer of the commodity has a purchase intention or a consultation demand, the buyer can initiate an interactive chat with the seller user through a consultation portal provided by the transaction platform and consult the seller about the detailed information of the commodity. After the buyer and the seller make speeches, the terminal device can obtain the consultation dialogue information corresponding to the commodity.

[0141] After obtaining the consultation dialogue information, the terminal device can perform question and answer information identification based on the consultation dialogue information to obtain the commodity description information of the commodity. In some embodiments, a question and answer identification model can be run on the terminal device, and the question and answer identification model can refer to the description in the foregoing embodiments, which will not be described here. The terminal device can perform question and answer information identification on the consultation dialogue information based on the question and answer model to obtain the description information of the commodity. In other embodiments, the terminal device can send the obtained consultation dialogue information to a server, and the server can perform question and answer information identification based on the question and answer identification model provided in the foregoing embodiments to obtain the description information of the commodity, and return the description information to the terminal device.

[0142] After the terminal device obtains the description information of the commodity, the terminal device can use the commodity description information as the display information of the commodity to display the commodity. The display information of the commodity refers to the information displayed in the introduction page of the commodity. In this implementation, the terminal device identifies the commodity description information from the commodity consultation dialogue information as the display information of the commodity, which can further enrich the information used to describe the commodity based on the content contained in the commodity consultation dialogue, and can also make full use of the existing commodity information resources to reduce the repeated consultation rate of commodity information. Based on the commodity information display method provided in the foregoing embodiments, in some cases, the transaction platform can independently perform question and answer information identification on the dialogue information between the buyer and the seller to obtain the commodity description information contained in the dialogue information. After obtaining the commodity description information, the transaction platform can independently display the commodity description information through the terminal device of the seller and recommend the seller to add the commodity description information to the display information of the commodity.

[0143] In some other cases, the seller can request the transaction platform to start the function of question and answer information recognition on the consultation dialogue information of the commodity through the terminal device. For example, the transaction platform can display an operation control through the terminal device, which can be displayed in the consultation dialogue interface between the seller and the buyer, or can be displayed in other interfaces, and the embodiment is not limited. The seller can trigger the operation control to request the transaction platform to start the function of question and answer information recognition. The transaction platform can respond to the request, and perform question and answer information recognition on the dialogue information between the buyer and the seller to obtain the commodity description information contained in the dialogue information. After obtaining the commodity description information, the transaction platform can display the commodity description information through the terminal device of the seller, and details are not repeated.

[0144] It should be noted that in some of the processes described in the above embodiments and the accompanying drawings, a plurality of operations are included in a specific order, but it should be clearly understood that these operations can be executed or in parallel in the order in which they appear in this document. The serial numbers of the operations, such as 801, 802, etc., are only used to distinguish different operations, and the serial numbers themselves do not represent any execution order. In addition, these processes can include more or fewer operations, and the operations can be executed in sequence or in parallel.

[0145] It should be noted that the "first", "second", and the like described herein are used to distinguish different messages, devices, modules, etc., and do not represent the order of precedence. "First" and "second" are different types.

[0146] The above describes optional embodiments of the question and answer information processing method applicable to the terminal device side, such as Figure 9 As shown in FIG. 9, the terminal device can include a memory 900, a processor 901, a communication component 902, and a display component 903.

[0147] The memory 900 can be configured to store other various data to support operations on the terminal device. Examples of these data include instructions for operating any application or method on the terminal device, contact data, phonebook data, messages, pictures, videos, etc. The memory can be realized by any type of volatile or non-volatile storage device or their combination, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk or optical disk.

[0148] In this embodiment, the memory 900 is used to store one or more computer instructions.

[0149] The processor 901, coupled to the memory 900, is configured to execute one or more computer instructions in the memory 900, to: receive object information of a target object sent by a server through a communication component 902; the object information is obtained according to at least one round of consultation dialogue between a consultation user and a response user; display the object information to the response user in a consultation dialogue interface of the target object through a display component 903; and in response to a confirmation operation of the response user on the object information, send an addition request to the server to add the object information to the detail information of the target object.

[0150] In some example embodiments, when the processor 901 displays the object information to the response user in the consultation dialogue interface of the target object, the processor 901 is specifically configured to: display a prompt message of supplementary information in the consultation dialogue interface of the target object through a display component 904; in response to a triggering operation on the prompt message, display a floating window; in the floating window, display a target attribute corresponding to the object information and a target attribute value; or in the floating window, display a question and answer pair corresponding to the object information.

[0151] In some example embodiments, when the processor 901 displays the question and answer pair corresponding to the object information in the floating window through the display component 903, the processor 901 is further configured to: display a modification control of the question and answer pair through the display component 903; and in response to a triggering operation on the modification control, obtain a modification result of the response user on the question and answer pair as an updated question and answer pair.

[0152] Further optionally, Figure 9 The schematic terminal device further includes an audio component 904 and a power component 905.

[0153] In an optional implementation, the display component 903 is configured to display a to-be-filled information page and an interaction result. The display component 903 includes a liquid crystal display (LCD) and a touch panel (TP). If the display component 903 includes the touch panel, the display component 903 can be implemented as a touch screen to receive an input signal from a user. The touch panel includes one or more touch sensors to sense touch, sliding and gestures on the touch panel. The touch sensor can not only sense the boundary of a touch or sliding action, but also detect the duration and pressure associated with the touch or sliding operation.

[0154] In an optional embodiment, the audio component 904 is configured to store output and / or input audio signals. For example, the audio component 904 includes a microphone (MIC) configured to receive an external audio signal when the device in which the audio component 904 is located is in an operational mode, such as a call mode, a recording mode, and a voice recognition mode. The received audio signal can be further stored in the memory 900 or transmitted via the communication component 902. In some embodiments, the audio component 904 also includes a speaker for outputting audio signals.

[0155] In an optional embodiment, the power component 905 is configured to provide power for various components of the terminal device. The power component can include a power management system, one or more power sources, and other components associated with generating, managing, and distributing power for the terminal device.

[0156] In this embodiment, in the scenario of a human-to-human consultation dialogue, the terminal device can receive object information identified by the server based on at least one round of the human-to-human consultation dialogue, and display the object information. The object information can be displayed in the detail information of the target object after being confirmed by the responding user. In this way, the depth of the dialogue information between the human-to-human is mined, and effective information can be mined from the dialogue information and reasonably utilized, which is conducive to reducing the time cost and labor cost required for repeated consultation and repeated consultation reply.

[0157] In addition to the question and answer information processing logic described in the foregoing embodiments, Figure 9 The terminal device shown can also execute the commodity information display logic. Specifically, the processor 901 is configured to: obtain consultation dialogue information corresponding to a commodity; perform question and answer information identification based on the consultation dialogue information to obtain commodity description information of the commodity; and display the commodity description information as display information of the commodity to display the commodity. In this embodiment, the terminal device can identify commodity description information from commodity consultation dialogue information, and display the commodity description information as display information of the commodity. In this way, on the one hand, the information used to describe the commodity can be further enriched based on the content contained in the commodity consultation dialogue, and on the other hand, the existing commodity information resources can be fully utilized to reduce the rate of repeated consultation of commodity information.

[0158] Correspondingly, the embodiments of the present application also provide a computer readable storage medium storing a computer program, which is executed to implement the steps of the method embodiments that can be executed by the terminal device.

[0159] The above describes optional embodiments of the question and answer information processing method applicable to the server side, such as Figure 10 As shown, in practice, the server can include a memory 1000, a processor 1001, and a communication component 1002.

[0160] The memory 1000 can be configured to store other various data to support operations on the server. Examples of such data include instructions for any application or method operating on the server, contact data, phonebook data, messages, pictures, videos, and the like. The memory can be realized by any type of volatile or nonvolatile storage devices or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic or optical disk.

[0161] In the embodiment, the memory 110 is configured to store one or more computer instructions.

[0162] The processor 1001, coupled to the memory 1000, is configured to execute one or more computer instructions in the memory 1000, so as to: acquire, by a communication component 1002, at least one round of consultation dialogue between a consultation user and a response user for a target object; perform question and answer information identification on the at least one round of consultation dialogue to obtain object information of the target object; send the identified object information to a terminal device of the response user, so that the terminal device displays the object information to the response user; and receive an addition request returned by the terminal device according to a confirmation operation of the response user on the object information, and add the object information to detail information of the target object.

[0163] In some exemplary embodiments, when acquiring the at least one round of consultation dialogue between the consultation user and the response user, the processor 1001 is specifically configured to: according to a user question of the consultation user, identify an interaction intention of the consultation user; and if the interaction intention is used to inquire attribute information of the target object, acquire, as the at least one round of consultation dialogue, a specified round of dialogue before each time a reply message of the response user is received.

[0164] In some example embodiments, the processor 1001 is specifically configured to, when performing question-answer information identification on the at least one round of consultation dialogue to obtain at least one object information of the target object, input the at least one round of consultation dialogue into a question-answer extraction model; perform word segmentation processing on a plurality of sentences contained in the at least one round of consultation dialogue respectively in the question-answer extraction model to obtain respective word segmentation results of the plurality of sentences; perform semantic coding on the respective word segmentation results of the plurality of sentences to obtain respective coding vectors of the plurality of sentences; perform fusion processing on the respective coding vectors of the plurality of sentences by using at least one of a speaking role, a question-answer position, and a speaking time of the plurality of sentences to obtain respective fusion vectors of the plurality of sentences; identify a question sentence and a reply sentence related to an attribute of the target object from the plurality of sentences according to the respective fusion vectors of the plurality of sentences; and obtain the at least one object information of the target object from the question sentence and the reply sentence related to the attribute of the target object.

[0165] In some example embodiments, the processor 1001 is specifically configured to, when performing fusion processing on the respective coding vectors of the plurality of sentences by using at least one of a speaking role, a question-answer position, and a speaking time of the plurality of sentences to obtain respective fusion vectors of the plurality of sentences, for any one of the plurality of sentences, splice, in the question-answer extraction model, a coding vector of the sentence, a speaking role corresponding to the sentence, and a keyword of the sentence to obtain a first feature vector of the sentence; perform position coding on the first feature vector and question-answer position information of the sentence in the at least one round of question-answer to obtain a second feature vector; input the second feature vector into a multi-head attention network to obtain a third feature vector of the sentence; and splice the third feature vector and speaking time information of the sentence to obtain a fourth feature vector as the fusion vector of the sentence.

[0166] In some example embodiments, the processor 1001 is specifically configured to, when obtaining the at least one object information of the target object from the question sentence and the reply sentence related to the attribute of the target object, perform question-answer pair matching on the question sentence and the reply sentence related to the attribute of the target object to obtain at least one group of question-answer pairs; determine an output mode of each of the at least one group of question-answer pairs according to semantic information of each of the at least one group of question-answer pairs; the output mode includes a structured output or an unstructured output; and output the at least one group of question-answer pairs as the at least one object information of the target object according to the output mode of each of the at least one group of question-answer pairs.

[0167] In some example embodiments, after matching the question sentences and the answer sentences for the target object to obtain at least one set of question-answer pairs, the processor 1001 is further configured to: for any one set of question-answer pairs, if the question-answer pair contains multiple question sentences, stitching the multiple question sentences in the question-answer pair; and if the question-answer pair contains multiple answer sentences, stitching the multiple answer sentences in the question-answer pair.

[0168] In some example embodiments, when determining the output mode of each of the at least one set of question-answer pairs according to the semantic information of each of the at least one set of question-answer pairs, the processor 1001 is specifically configured to: for any one set of question-answer pairs in the at least one set of question-answer pairs, performing semantic recognition on the question-answer pair to obtain the semantic of the question-answer pair; if there is an attribute in the pre-set structured attribute of the target object that matches the semantic of the question-answer pair, determining to output the question-answer pair in the structured output mode; and if there is no attribute in the pre-set structured attribute of the target object that matches the semantic of the question-answer pair, determining to output the question-answer pair in the unstructured output mode.

[0169] In some example embodiments, when outputting the at least one set of question-answer pairs as at least one object information of the target object according to the output mode of each of the at least one set of question-answer pairs, the processor 1001 is specifically configured to: if the question-answer pair is output in the structured output mode, performing attribute extraction on the question-answer pair to obtain a target attribute and a corresponding target attribute value of the question-answer pair; forming structured information of the target attribute and the target attribute value as the object information corresponding to the question-answer pair; and if the question-answer pair is output in the unstructured output mode, performing information security processing on the question-answer pair, and taking the question-answer pair obtained by the information security processing as the object information corresponding to the question-answer pair.

[0170] In some example embodiments, after forming the structured information of the target attribute and the target attribute value as the object information corresponding to the question-answer pair, the processor 1001 is further configured to: querying an existing attribute value of the target attribute from the attributes already displayed in the detail page of the target object; and if the existing attribute value of the target attribute is different from the target attribute value, returning information prompting to modify the attribute value of the target attribute to the terminal device.

[0171] In some example embodiments, after obtaining the object information corresponding to the question-answer pair by processing the information securely, the processor 1001 is further configured to: determine a target historical question identical to the question in the question-answer pair from historical questions that have been displayed in the detail page of the target object; determine whether the semantic of the answer of the target historical question is identical to the semantic of the answer of the question-answer pair; and if yes, send the answer of the question-answer pair as the answer of the target historical question to the terminal device.

[0172] Further optionally, Figure 10 The server also includes a power supply component 1003. The power supply component 1003 is configured to provide power to various components of the server. The power supply component can include a power management system, one or more power supplies, and other components associated with generating, managing, and distributing power to the server.

[0173] In the present embodiment, in the scenario of a human-to-human counseling dialogue, the question-answer recognition can be performed according to at least one round of counseling dialogue between the human and the human for the target object, so as to obtain the object information of the target object. The object information can be displayed in the detail information of the target object after being confirmed by the responding user. Based on this manner, the deep mining of the dialogue information between the human and the human is realized, the effective information can be mined from the dialogue information and reasonably utilized, and the time cost and the labor cost required for repeated counseling and repeated counseling reply are reduced.

[0174] Correspondingly, the embodiments of the present application also provide a computer readable storage medium storing a computer program, which is executed to implement each step in the method embodiments that can be executed by the human server in the method embodiments.

[0175] Those skilled in the art should understand that the embodiments of the present application can be provided as a method, a system, or a computer program product. Therefore, the present application can take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present application can take the form of a computer program product implemented on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROMs, optical storage media, etc.) containing computer-usable program code.

[0176] The computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart block or blocks. Figure 1 one or more flowcharts and / or blocks in the flowcharts and / or combination thereof. Figure 1 one or more flowcharts and / or blocks in the flowcharts and / or combination thereof.

[0177] The computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart block or blocks. Figure 1 one or more flowcharts and / or blocks in the flowcharts and / or combination thereof. Figure 1 one or more flowcharts and / or blocks in the flowcharts and / or combination thereof.

[0178] The computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart block or blocks. Figure 1 one or more flowcharts and / or blocks in the flowcharts and / or combination thereof. Figure 1 one or more flowcharts and / or blocks in the flowcharts and / or combination thereof.

[0179] In one typical configuration, the computing device includes one or more processors (CPUs), input / output interfaces, network interfaces, and memory.

[0180] The memory can include non-persistent memory and / or volatile memory, such as random access memory (RAM) and / or cache memory, non-volatile memory, such as read-only memory (ROM), EPROM, and / or flash memory. The memory is an example of computer-readable media.

[0181] Computer-readable media includes permanent and non-permanent, movable and non-movable media that can implement information storage by any method or technology. The information can be computer-readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, compact disc read-only memory (CD-ROM), digital versatile disc (DVD) or other optical storage, magnetic cassette, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other non-transmission medium that can be used to store information accessible to a computing device. According to the definition herein, computer-readable media does not include transitory media such as modulated data signals and carriers.

[0182] It should also be noted that the terms "comprising", "containing", or any other variant thereof are intended to cover a non-exclusive inclusion, such that a process, method, article or apparatus that comprises a list of elements does not only include those elements, but can also include other elements not expressly listed or inherent to such process, method, article or apparatus. Without more limitations, the element defined by the statement "comprising a" does not exclude the presence of additional identical elements in the process, method, article or apparatus that includes the element.

[0183] The above only describes the embodiments of the present application and is not intended to limit the present application. For those skilled in the art, the present application can have various modifications and changes. Any modification, equivalent replacement, improvement, etc. within the spirit and principle of the present application shall be included in the scope of claims of the present application.

Claims

1. A question-and-answer information processing method, characterized in that, include: Receive object information of the target object sent by the server; The object information is obtained by identifying question and answer information from at least one round of consultation dialogue between the consulting user and the responding user regarding the target object. The question-and-answer information recognition is initiated when the interaction intent of the consulting user is detected as being used to inquire about the attribute information of the target object; The question-and-answer information recognition includes: extracting statements for asking questions about the attributes of the target object and statements for answering questions about the attributes from the at least one round of consultation dialogue, and obtaining the object information based on the question statements and answer statements; and displaying the object information to the responding user in the consultation dialogue interface of the target object. In response to the user's confirmation of the object information, an add request is sent to the server so that the server adds the object information to the details information of the target object; the object information is used to be displayed in the details information when the consulting user, the responding user, or other users request to view the details information of the target object, and the details information includes the product details page corresponding to the target object.

2. The method according to claim 1, characterized in that, In the consultation dialogue interface of the target object, the object information is displayed to the responding user, including: In the consultation dialogue interface of the target object, a prompt message for supplementary information is displayed; In response to the triggering operation of the prompt message, a floating window is displayed; The floating window displays the target attributes and target attribute values ​​corresponding to the object information; or, the floating window displays the question-and-answer pairs corresponding to the object information.

3. The method according to claim 2, characterized in that, In the consultation dialogue interface of the target object, displaying the object information to the responding user also includes: Display the modification controls corresponding to the object information; In response to the trigger operation of the modification control, the modification result of the responding user on the object information is obtained and used as the updated object information.

4. A method for displaying product information, characterized in that, include: Obtain consultation and dialogue information corresponding to the product; Based on the consultation dialogue information, question and answer information is identified to obtain the product description information of the product. The question and answer information identification is initiated when the interaction intent of the consulting user is detected to be asking for the attribute information of the product. The question-and-answer information recognition includes: extracting statements for asking about the attributes of the product and statements for answering about the attributes from the consultation dialogue information, and obtaining the product description information based on the question statements and answer statements; The product description information is used as the product display information to display the product; the product description information is used to display the product description information on the product's introduction page when the consulting user, responding user, or other user corresponding to the consultation dialogue information requests to view the product display information.

5. A question-and-answer information processing method, characterized in that, include: Acquire at least one round of consultation dialogue between the consulting user and the responding user regarding the target audience; The process of identifying question-and-answer information in the at least one round of consultation dialogue to obtain object information of the target object includes: extracting statements for asking questions about the attributes of the target object and statements for answering questions about the attributes from the at least one round of consultation dialogue, and obtaining the object information based on the question statements and answer statements. The identified object information is sent to the terminal device of the responding user, so that the terminal device displays the object information to the responding user; The terminal device receives an add request returned by the responding user based on the confirmation operation of the object information, and adds the object information to the details information of the target object; the object information is used to display in the details information when the consulting user, the responding user, or other user requests to view the details information of the target object, and the details information includes: the product details page corresponding to the target object.

6. The method according to claim 5, characterized in that, Acquiring at least one round of consultation dialogue between the consulting user and the responding user, including: Identify the user's interaction intent based on the user's question; If the interaction intent is used to inquire about the attribute information of the target object, then each time a reply message from the responding user is received, a specified number of rounds of dialogue prior to the reply message are obtained as the at least one round of consultation dialogue.

7. The method according to claim 5, characterized in that, Question-and-answer information recognition is performed on the at least one round of consultation dialogue to obtain at least one type of object information of the target object, including: Input the at least one round of consultation dialogue into the question-and-answer extraction model; In the question-and-answer extraction model, the multiple statements contained in the at least one round of consultation dialogue are segmented into words to obtain the segmentation results of each of the multiple statements. Semantic encoding is performed on the word segmentation results of the multiple statements to obtain the encoding vectors of the multiple statements; Using at least one of the speaking roles, question-and-answer positions, and speaking times of the multiple statements, the encoding vectors of the multiple statements are fused to obtain the fused vectors of the multiple statements. Based on the fusion vectors of the multiple statements, question statements and answer statements related to the attributes of the target object are identified from the multiple statements; Obtain at least one type of object information of the target object from the question and answer statements related to the attributes of the target object.

8. The method according to claim 7, characterized in that, Using at least one of the following information—the speaker's role, question-and-answer position, and speaking time—the encoding vectors of the multiple statements are fused to obtain the fused vectors of the multiple statements, including: For any one of the plurality of statements, in the question-and-answer extraction model, the encoding vector of the statement, the speaking role corresponding to the statement, and the keywords of the statement are concatenated to obtain the first feature vector of the statement; The first feature vector is positionally encoded with the question-and-answer position information of the statement in at least one round of question and answer to obtain the second feature vector; The second feature vector is input into a multi-head attention network to obtain the third feature vector of the statement; The third feature vector is concatenated with the speaking time information of the statement to obtain the fourth feature vector, which is used as the fusion vector of the statement.

9. The method according to claim 7, characterized in that, Obtaining at least one type of object information about the target object from query statements and response statements related to the attributes of the target object, including: Match the question and answer statements related to the attributes of the target object to obtain at least one set of question and answer pairs; Based on the semantic information of each of the at least one set of question-answer pairs, determine the output method of each of the at least one set of question-answer pairs; the output method includes: structured output or unstructured output; Based on the respective output methods of the at least one set of question-and-answer pairs, the at least one set of question-and-answer pairs is output as at least one type of object information of the target object.

10. The method according to claim 9, characterized in that, Based on the semantic information of each of the at least one set of question-answer pairs, determine the output method of each of the at least one set of question-answer pairs, including: For any one of the at least one set of question-answer pairs, perform semantic recognition on the question-answer pairs to obtain the semantics of the question-answer pairs; If the target object has a preset structured attribute that semantically matches the question-answer pair, then the structured output method is used to output the question-answer pair. If the target object does not have any preset structured attributes that semantically match the question-answer pair, then the question-answer pair will be output using the unstructured output method.

11. The method according to claim 10, characterized in that, Based on the respective output methods of the at least one set of question-and-answer pairs, the at least one set of question-and-answer pairs is output as at least one type of object information of the target object, including: If the question-and-answer pair is output using the structured output method, then attribute extraction is performed on the question-and-answer pair to obtain the target attribute and the corresponding target attribute value of the question-and-answer pair; the structured information formed by the target attribute and the target attribute value is used as the object information corresponding to the question-and-answer pair. If the question-and-answer pair is output using the unstructured output method, then the question-and-answer pair is subjected to information security processing, and the question-and-answer pair obtained from the information security processing is used as the object information corresponding to the question-and-answer pair.

12. A terminal device, characterized in that, include: Memory, processor, communication components, and display components; The memory is used to store one or more computer instructions; The processor is configured to execute one or more computer instructions for performing the steps of the method according to any one of claims 1-4.

13. A server, characterized in that, include: Memory, processor, and communication components; The memory is used to store one or more computer instructions; The processor is configured to execute one or more computer instructions for performing the steps of the method according to any one of claims 5-11.

14. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed, it can perform the steps of the method according to any one of claims 1-4 or the steps of the method according to any one of claims 5-11.

Images