WO2019144858A1 - 人机对话方法及电子设备 - Google Patents

人机对话方法及电子设备 Download PDF

Info

Publication number
WO2019144858A1
WO2019144858A1 PCT/CN2019/072603 CN2019072603W WO2019144858A1 WO 2019144858 A1 WO2019144858 A1 WO 2019144858A1 CN 2019072603 W CN2019072603 W CN 2019072603W WO 2019144858 A1 WO2019144858 A1 WO 2019144858A1
Authority
WO
WIPO (PCT)
Prior art keywords
jump
topic
topics
user
initial
Prior art date
Application number
PCT/CN2019/072603
Other languages
English (en)
French (fr)
Inventor
初敏
郭涛涛
戴中原
杨超
Original Assignee
苏州思必驰信息科技有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 苏州思必驰信息科技有限公司 filed Critical 苏州思必驰信息科技有限公司
Priority to JP2020537003A priority Critical patent/JP7288446B2/ja
Priority to US16/769,272 priority patent/US11735163B2/en
Priority to EP19743677.7A priority patent/EP3745283A4/en
Publication of WO2019144858A1 publication Critical patent/WO2019144858A1/zh

Links

Images

Classifications

    • G—PHYSICS
    • G10—MUSICAL INSTRUMENTS; ACOUSTICS
    • G10L—SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
    • G10L15/00—Speech recognition
    • G10L15/02—Feature extraction for speech recognition; Selection of recognition unit
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
    • G06F16/33—Querying
    • G06F16/335—Filtering based on additional data, e.g. user or group profiles
    • G06F16/337—Profile generation, learning or modification
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
    • G06F16/33—Querying
    • G06F16/332—Query formulation
    • G06F16/3329—Natural language query formulation
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
    • G06F16/36—Creation of semantic tools, e.g. ontology or thesauri
    • G06F16/367—Ontology
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N5/00—Computing arrangements using knowledge-based models
    • G06N5/04—Inference or reasoning models
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N7/00—Computing arrangements based on specific mathematical models
    • G06N7/01—Probabilistic graphical models, e.g. probabilistic networks
    • G—PHYSICS
    • G10—MUSICAL INSTRUMENTS; ACOUSTICS
    • G10L—SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
    • G10L15/00—Speech recognition
    • G10L15/22—Procedures used during a speech recognition process, e.g. man-machine dialogue

Definitions

  • the present invention relates to the field of artificial intelligence technologies, and in particular, to a human-machine dialogue method and an electronic device.
  • Task-based conversations Vertical-field conversations tailored to the tasks that users often need: ordering, booking, finding music, movies, or a certain item. After the user speaks a sentence, he first determines which task is required and extracts the user's demand parameters (eg, departure place, restaurant type, etc.). If the necessary parameters defined in advance are not collected, the machine will ask for information by asking questions. Therefore, task-based conversations are usually multiple rounds of dialogue. Users may also continually modify or refine their needs during the conversation.
  • demand parameters eg, departure place, restaurant type, etc.
  • Dialogue based on question and answer Knowledge is organized in a question-and-answer format, comparing the user's questions with the questions in the question and answer pairs, finding the closest one, and returning the answer. This kind of dialogue is often used for customer service robots and gossip robots. Most of these Q&A services are single-disciplinary, and some have some multi-round dialogue skills, mainly involving simple context processing and referencing disambiguation.
  • Knowledge Graph-based Dialogue Users use natural language/speech to query factual knowledge stored in triples. For example, "How tall is Yao Ming's daughter?" In this kind of dialogue, the robot needs certain reasoning ability. The above sentence is actually completed in two steps (1) Yao Ming's daughter -> Yao Yulei; (2) Yao Yulei's height -> 160cm. Most of these conversations are single-round, and some have a certain number of dialogues, mainly referring to disambiguation (she is a few years old? She-> Yao Yulei).
  • Generating Chat By training a neural network model, you can automatically generate a response based on a user question. This type of chat does not have clear communication goals and domain limits. When the user speaks a sentence, the system automatically generates a reply, and the response has a certain relationship with the problem, but there is no clear communication goal. It is also said that such conversations are open domain chats. In the existing man-machine dialogue system, open domain chat mainly plays the role of close distance, establishing trust relationship, emotional companionship, smooth dialogue process (for example, when task-type dialogue cannot meet user needs) and improving user stickiness. .
  • Some of the various intelligent robots in the prior art use one of them, and some are combinations of several forms.
  • machines In all these forms of dialogue, basically people are active parties, machines are passive, machines wait for people to ask questions, and then give answers.
  • the machine In a task-based conversation, the machine also asks questions, but in a very well-defined situation. For example, to buy a ticket, define three required parameters in advance: departure location, arrival location, and departure time. Only when the user tells the whole, the robot can issue a query to the ticket service system. If not, the robot will ask for the missing information. But overall, the machine is still passively waiting.
  • various question and answer robots are passive at present, and the robot is waiting for the user to ask questions. After the user actively asks the question, the robot begins to understand the user's intention, and queries and feedbacks the answers according to the user's intention, or performs actions. Although there are multiple rounds of dialogue, it is mainly to ask the user some necessary parameters (what day is it going? What color do you like?). The robot itself has no clear intention of dialogue. In such a dialogue system, if the user can't think of the topic or the specific way of asking questions, the dialogue can't go on, and the user can't think about what the user thinks and provide a better interactive experience for the user.
  • the embodiment of the invention provides a human-machine dialogue method and an electronic device for solving at least one of the above technical problems.
  • an embodiment of the present invention provides a human-machine dialog method, which is applied to an electronic device, where the method includes:
  • the jump topic to which the user's dialog request belongs is selected from the topic jump map as an initial topic for performing the first round of recommendation to the user;
  • the initial topic is gradually guided from the initial topic to the target topic in a stepwise recommendation manner.
  • an embodiment of the present invention provides another human-machine dialog method, which is applied to an electronic device, and the method includes:
  • the referral topic in the topic jump map is continuously recommended in a round-by-round recommendation manner for human-machine dialogue.
  • an embodiment of the present invention provides a non-transitory computer readable storage medium, where the storage medium stores one or more programs including execution instructions, which can be used by an electronic device (including but not It is limited to a computer, a server, or a network device, etc., and is read and executed for performing the above-described human-machine dialog method of the present invention.
  • an electronic device including but not It is limited to a computer, a server, or a network device, etc.
  • an electronic device comprising: at least one processor, and a memory communicatively coupled to the at least one processor, wherein the memory stores instructions executable by the at least one processor, The instructions are executed by the at least one processor to enable the at least one processor to perform any of the above-described human-machine dialog methods of the present invention.
  • an embodiment of the present invention further provides a computer program product, the computer program product comprising a computer program stored on a non-transitory computer readable storage medium, the computer program comprising program instructions, when the program When the instruction is executed by the computer, the computer is caused to execute any of the above-described human-machine dialog methods.
  • the beneficial effects of the embodiment of the present invention are: by using a topic jump map established based on the association strength between jump topics in advance, and determining the first conversation request of the interlocutor in the human-machine dialogue process as the corresponding topic jump
  • the jump topic in Tup and then actively recommend the next jump topic related to the topic currently discussed by the user according to the strength of the association between the jump topics, so that the computer answers the question raised by the user, or based on
  • the user asks the user to make recommendations on related topics, and can always be in an efficient operation state, which improves the utilization rate of the computer during the human-machine dialogue process.
  • the dialogue between the person and the machine is ensured smoothly, and the user is improved.
  • FIG. 1 is a flow chart of an embodiment of a human-machine dialog method according to the present invention.
  • FIG. 2 is a flow chart of another embodiment of a human-machine dialog method according to the present invention.
  • FIG. 3 is a flow chart of still another embodiment of a human-machine dialog method according to the present invention.
  • FIG. 4 is a flowchart of still another embodiment of a human-machine dialog method according to the present invention.
  • FIG. 5 is a schematic structural view of an embodiment of an electronic device according to the present invention.
  • the invention may be described in the general context of computer-executable instructions executed by a computer, such as a program module.
  • program modules include routines, programs, objects, elements, data structures, and the like that perform particular tasks or implement particular abstract data types.
  • the invention may also be practiced in distributed computing environments where tasks are performed by remote processing devices that are connected through a communication network.
  • program modules can be located in both local and remote computer storage media including storage devices.
  • module refers to related entities applied to a computer, such as hardware, a combination of hardware and software, software or software in execution, and the like.
  • an element can be, but is not limited to being, a process running on a processor, a processor, an object, an executable element, a thread of execution, a program, and/or a computer.
  • an application or script running on a server, or a server can be a component.
  • One or more elements can be executed in a process and/or thread, and the elements can be localized on a computer and/or distributed between two or more computers and can be executed by various computer readable media.
  • the component may also pass signals based on data having one or more data packets, for example, from a signal interacting with another element in the local system, the distributed system, and/or interacting with other systems over the network of the Internet. Local and/or remote processes to communicate.
  • an embodiment of the present invention provides a topic jump map optimization method for a robot dialogue, including:
  • the target topic in the embodiment of the present invention may be at least one topic determined in advance, and a series of (for example, 50-100) topics related to the target topic may be used to generate a topic jump map, and the determined number of jumps may be determined.
  • Topics are categorized, for example, into: key topics (some of which may be important to achieve goals, try to make users aware), related topics (some supplements to key topics, help to achieve communication goals) and interesting topics (Increase the topic of communication fun).
  • Each of the jump topics and/or the target topics respectively contains a plurality of knowledge points, the knowledge points being represented by a question and answer pair and/or by a knowledge map.
  • the strength of the association between the jump topics is determined according to various strategies such as the analogy relationship of the jump topic and the reference relationship (after a series of topics are obtained, the topic jump map is constructed according to the relationship between the topics. For example, two conceptually exist. A jump relationship is established between the topics of the father and son, the inclusion, the parallel, the reference, or the analogy relationship; the jump relationship may also be set according to the semantic similarity of the two sets of knowledge points included in the topic; and the literature may be based on related topics in the literature, books, The strength of the associated association in the network information is used to establish a jump relationship, which is not limited by the present invention.
  • a topic path that can be jumped next is established, and an initial probability is assigned to each jump path (per The initial probabilities of a path).
  • the initial probabilities of a path may be the same or different. For example, different types of relationships correspond to different initial probabilities, or corresponding initial probabilities are calculated according to the size of semantic similarity, the strength of association between topics, and the like. If industry experts have special needs, you can also specify jump paths and probabilities.
  • After topic planning we get a finite state machine between topics. Each topic represents a state, and each topic jump is a directed edge with probability. In order to ensure that the dialogue process has a certain diversity, and at the same time, it can effectively achieve the goal of dialogue.
  • the jump target is set to about 10% of the total number of topics for a topic (the actual application is adjusted according to the total amount of topics).
  • downstream level described in the embodiment of the present invention is only related to the current jump topic, and there is no relationship between the upstream and downstream levels in the entire topic jump map, for example, for the current topic A1. If the topic that topic A1 can jump to includes topics B, C, and D, then topic B, C, and D are the jump topics of the downstream level of topic A1, if topic A and topic A2 still exist, and topic A1. A2 is the downstream level of topic A, respectively, but topic A2 cannot jump to topics B, C, and D (ie, there is no jump path between these topics).
  • the user asks “What products does your company have?”, then understands the natural language and returns the answer “XXX company focuses on intelligent conversational interaction in the vertical field.
  • the main products include smart car solutions, smart home solutions, and intelligence. Internet of things such as robotics and the field of Internet of Things.”
  • the topic that contains the user's question is determined as the initial topic, specifically, the statement containing the same or similar to the user's question (included in the dialogue pair), or the knowledge map includes the knowledge points associated with the user's question.
  • the jump topic downstream from the initial topic is determined based on the topic jump map, and the target jump topic is selected from the downstream jump topic according to the jump probability, and the target jump topic is determined from the determined target
  • the selection question included in the included knowledge points is recommended to the user for human-machine dialogue. After completing the dialogue of the current target jump topic, further determining the downstream jump node to which the current target jump topic can jump to... and so on until determining The entire jump path from the initial jump topic to the target topic.
  • selecting a target jump topic from the downstream jump topic according to the jump probability may select the target topic according to the probability sampling. For example, if topic A has three jump paths that jump to topics B, C, and D, respectively, and the corresponding jump probability is 0.5, 0.3, and 0.2, then sampling by probability means that if 100 human-machine interactions are performed, 50 Take the first path (AB), go to the second (AC) 30 times, and go to the third (AD) 20 times.
  • the topic jump map in the robot dialog method of the embodiment of the present invention is obtained based on a plurality of jump topic plans around the target topic, and the jump map finally jumps to the target topic, thereby causing the robot of the embodiment of the present invention
  • the dialogue can have a certain dialogue intention (target topic), and can gradually jump the final topic to the target topic of the dialogue intention by jumping to the target topic jump convergence peripheral jump topic, thereby facilitating the dialogue between the user and the robot. Go smoothly, and don't talk about the topic (off topic, that is, off target).
  • the robot since the method of recommending the downstream jump topic by level is adopted, the robot can be made to have a certain degree of initiative in the dialogue process, so that on the one hand, the dialogue can be smoothly carried out and finally jump to the target topic to complete the dialogue purpose.
  • the recommendation process because it is a step-by-step recommendation, it is not directly blunt to enter the target topic, which will not cause the user's dislike, and also enhance the user experience.
  • the human-machine dialogue method in the embodiment of the present invention can be embodied in various product forms, for example, "sales assistant", "business card”, and the like.
  • the sales assistant the purpose of achieving the sales target can be achieved, so it is possible to determine topics such as “inquiry”, “order”, “after-sale” as the target topic, and determine the surrounding topic around the product to be sold as a jump.
  • the topic generates a topic jump map that converges to the target topic based on the strength of the association between the jump topics, thereby being used for conducting a human-machine dialogue for sales purposes.
  • the "business card” can be positioned to spread the key information of the enterprise or achieve cooperation as a target function.
  • the human-machine dialogue method of the present invention can promote the promotion and product sales of the enterprise in the process of providing a friendly interactive experience for the user.
  • the sales assistant, the business card, and the like can be expressed as an application or a WeChat applet, etc., which is not limited by the present invention.
  • a topic related to sales for example, an inquiry, an order, an after-sales, etc.
  • a topic for example, a company profile, a case, a technical principle, and the like
  • Turn the topic to construct a topic jump map.
  • the user can click to open the sales assistant application, and display multiple navigation labels (for example, company profile, case, technical principle, purchase, etc.) under the display interface of the application, and the user can use voice input. Select the navigation tag of interest, and start the man-machine dialogue from the topic corresponding to the navigation tag selected by the user, and gradually lead the topic to the target topic based on the constructed topic jump map.
  • multiple navigation labels for example, company profile, case, technical principle, purchase, etc.
  • a jump relationship from the jump topic to the target topic may be: company profile ⁇ case ⁇ technical principle ⁇ purchase.
  • human-machine dialogue give an example of the following human-machine dialogue:
  • Sales Assistant: XXX Company was founded in X years, focusing on intelligent conversational interaction in the vertical field.
  • the main products include intelligent vehicle solutions, smart home solutions, intelligent robots and other Internet of Things and the Internet of Things.
  • Sales Assistant The company's voice assistant products have been applied to XX smart speakers, and microphone array products have been applied to car speakers of XX cars.
  • Sales Assistant The company's dual-use array is also used in...
  • the entire human-machine dialogue process starts from the topic selected by the user, and gradually leads the topic to the target topic (sale topic) in response to the user and based on the topic jump map recommendation to ask the user.
  • the human-machine dialog method further includes:
  • the completion of the human-machine dialogue in the embodiment of the present invention refers to that the user is satisfied with the human-machine dialogue, and can select the evaluation method before the end of the human-machine dialogue or answer the question that the robot actively proposes whether to be satisfied with the dialogue. It is determined that if the user expresses satisfaction with the human-machine dialogue, it is determined that the human-machine dialogue is completed and the subsequent steps are performed.
  • step S21 Since it has been determined in step S21 that the human-machine dialogue has been completed satisfactorily, this indicates that the jump topic that is recommended to the user and guided to the target topic during the human-machine dialogue process is more user-friendly.
  • these jump topics that are pointed to the target topic and recognized by the user are referred to. Set a larger jump probability.
  • the human-machine dialogue is carried out, and the continuous optimization of the topic jump map is realized, so that the next performance of the human-machine dialogue can have better performance, the response speed is faster and more accurate, and the user is improved.
  • the continuous optimization of the topic jump map is realized, so that the next performance of the human-machine dialogue can have better performance, the response speed is faster and more accurate, and the user is improved.
  • the amount of calculation of the computer can be reduced when the man-machine dialogue is performed again.
  • the computer may need to calculate the 10-step jump probability to achieve the target topic.
  • the optimization computer only needs to calculate the 5-step jump probability to achieve the target topic, which reduces the calculation amount of the computer and improves the target topic. effectiveness.
  • the sum of the initial jump probabilities of the k path segments of each of the k jump topics to which the jump topic can jump to the downstream level is 1;
  • the first jump probability may be represented by P f1
  • the n-1 second jump probabilities may be represented by P′ f1 —P′ f(n ⁇ 1) .
  • the sum of the jump probabilities of the path segments between the same jump topic and the k jump topics of the downstream level follows the normalization principle, and the stability of the entire topic jump map can be maintained, and Optimize management of topic jump maps for better human-machine dialogue.
  • the embodiment of the present invention further provides another human-machine dialogue method, including:
  • S31 Determine a set number of jump topics related to the specified domain, and construct a topic jump map between the topics based on the association strength between the set number of jump topics, where each jump topic is
  • the k path segments of the k jump topics of the downstream level that can be jumped to are configured with an initial jump probability; the designated fields in the embodiments of the present invention may be fields such as teaching, training, marketing, and the like.
  • the user By jumping the map based on the topic established in advance based on the strength of the association between the jump topics, and determining the first conversation request of the interlocutor in the human-machine dialogue process as the jump topic in the corresponding topic jump graph, after Then, according to the strength of the association between the jump topics, the user is actively recommended the next jump topic related to the topic currently discussed by the user, thereby ensuring the smooth progress of the dialogue between the person and the machine, and improving the user in the man-machine dialogue.
  • the topic discussed in the human-machine dialogue process can be guaranteed. It is always a topic of interest to users, so that users can talk about many topics with the robot without knowing it, and acquire multi-dimensional and multi-dimensional knowledge.
  • the human-machine dialogue method in the embodiment of the present invention can be embodied in various product forms, for example, “enterprise promotion business card”, “story machine”, “knowledge encyclopedia” and the like. Taking “enterprise promotion business card” as an example, you can determine multiple aspects related to the enterprise based on multiple dimensions of the enterprise (for example, enterprise development history, corporate culture, enterprise structure composition, enterprise advantage, enterprise products, enterprise talents, enterprise cases, etc.).
  • the topic is used to construct a topic jump map in the embodiment of the present invention, thereby implementing a human-machine dialogue mode capable of actively recommending a topic based on the map, and enabling the user to fully understand the enterprise in a friendly human-machine dialogue process.
  • the business promotion business card can be expressed as an application or a WeChat applet, etc., which is not limited by the present invention.
  • a topic jump map is constructed with associations between topics such as company profiles, cases, technical principles, products, product purchases, and the like.
  • the user can click to open the corporate promotional business card program, and display multiple navigation labels (for example, company profile, case, technical principle, product, product purchase, etc.) under the display interface of the application, and the user can use the voice.
  • the input method selects the navigation tag of interest, and starts the human-machine dialogue from the topic corresponding to the navigation tag selected by the user, and propagates the company information to the user based on the constructed topic jump map.
  • the jump relationship between topics in the topic jump map may be: company profile ⁇ product ⁇ technical principle ⁇ case ⁇ product ⁇ technical principle ⁇ case ⁇ product purchase ⁇ product.
  • human-machine dialogue give an example of the following human-machine dialogue:
  • XXX company was founded in X years, focusing on intelligent conversational interaction in the vertical field.
  • the main products include intelligent vehicle solutions, smart home solutions, intelligent robots and other Internet of Things and the Internet of Things.
  • corporate promotional business cards include: intelligent in-vehicle solutions, smart home solutions, intelligent robot solutions and voice input boards.
  • various topic information related to the current enterprise can be transmitted to the user through the topic jump map composed of different topics.
  • the jump connection relationship between the topics in the topic jump map is determined according to the association between the topics, each topic recommended to the user is a topic of interest to the user, thereby causing the computer to Is to answer the user's initiative to ask questions, or based on user questions to recommend relevant topics to the user, can always be in an efficient state of operation, improving the utilization of computers in the human-machine dialogue process.
  • the sum of the initial jump probabilities of the k path segments of each of the k jump topics to which the jump topic can jump to the downstream level is 1;
  • the process of referring to topics also includes:
  • the jump probability of the recommended jump topic is correspondingly corrected, so that the topic jump map can be optimized in time for subsequent implementation.
  • Human-machine dialogue services provide a more friendly dialogue guarantee.
  • the amount of calculation of the computer can be reduced when the man-machine dialogue is performed again.
  • a business card dialogue robot which wants to disseminate key information of the enterprise, and adopts the optimized topic jump map, can make the jump topic recommended by the computer to the user every time is a topic of interest to the user (there is no or at least reduced the computer) It is recommended that the user does not like the topic, but the computer needs to recalculate and recommend a new topic to the user), so that the computer can save a large amount of repetitive calculation of the repeated recommendation workload, and ensure the efficient use of computer processing power.
  • it can make the computer need to calculate the 10-step jump to achieve the purpose of disseminating the key information of the enterprise.
  • the optimized computer only needs to calculate the 5-step jump to achieve the purpose of disseminating the key information of the enterprise, reducing the calculation amount of the computer and improving the calculation. The efficiency of reaching the target topic.
  • the method further includes:
  • the topic independent of the first jump topic is reselected.
  • the predetermined number of times in the embodiment of the present invention may be 3-5 times, and the number of times the user recommends the jump topic is limited to a predetermined number of times, thereby avoiding excessive harassment caused by the topic content not recommended by the user, and
  • the jump probability of jumping to the jump topic rejected by the user is smaller than the jump probability of the initial jump probability, so that in the subsequent application to the optimization, the probability that the jump topic once rejected by the user is recommended again is lowered. To avoid causing trouble to the user again, thereby enhancing the friendliness of human-machine dialogue and further enhancing the user experience.
  • the at least one path segment in the topic jump map is disconnected, and the The topic jumps the map for the next human-machine conversation.
  • the optimization of the topic jump map is implemented in the human-machine dialogue process, if the jump probability value of the path segment of the jump topic to another jump topic is less than a predetermined threshold (such a path segment exists) At least one), directly delete the certain jump topic, so as to avoid the low probability jump path causing interference to the human-machine dialogue in the future, and on the other hand, it can also simplify the topic jump map, thereby accelerating the real-time effect of the human-machine dialogue.
  • a new jump topic is configured for a jump topic of an endpoint of the at least one path segment when the at least one path segment in the topic jump map is disconnected.
  • a plurality of jump topics that can be jumped for each current topic are selected from all topics associated with the current topic, The selected related topic is stored as a candidate jump topic set of the corresponding current topic.
  • a new jump topic configured for the jump topic of the endpoint of the at least one path segment is selected from the path segment.
  • a new jump configured for the jump topic of the endpoint of the at least one path segment
  • the topic may be a jump topic randomly selected from the topic jump map and given a jump probability.
  • the colleague who deletes the downstream jump topic in the at least one path segment whose jump probability is lower than the predetermined threshold also introduces a new jump topic, thereby ensuring the unreasonable jump map.
  • Optimized delete the jump topic that is frequently rejected by the user
  • ensures that the user is recommended to jump the diversity of the topic to ensure that the number of target jump topics that the current jump topic can jump to does not exceed Less
  • the method further includes: classifying the user according to the user attribute to generate a corresponding topic jump map according to the user's category.
  • the jump map of the corresponding category is generated for the users of different categories, so that the robot can be more suitable for the robot when the user uses the jump map robot in the embodiment to perform the dialogue.
  • the working mode can make the human-machine dialogue process smoother and enhance the user experience.
  • the topic of interest is different. For example, for a sales assistant robot, it needs to provide the user with an introduction to the company's various related information.
  • the consulting user is a purchasing person
  • the problem of concern may be more concentrated on the cost performance of the product, so that the jump map optimized for such a group of people must be based on the price/performance aspect of the product.
  • the user of the consultation is a technical research and development personnel
  • the concerns may be more focused on the introduction of topics on the performance principle of the product, so that the jump map constructed for such a group is necessarily a product.
  • the topic of performance is mainly topical.
  • the user attribute may include the user's job title, the user's gender, the user's age, and the like, and may be obtained by the user to fill in the above information before the user starts the consultation, or may be any other manner, which is not limited by the present invention.
  • the initial state of the conversation is determined according to the user behavior. If the user asks directly, the semantic understanding of the question is made, and the relevant knowledge points are found, and an accurate answer is given. After this:
  • Topic recommendation mode recommend one or more target jump topics directly to the user, and the user selects by clicking (this mode is only applicable to the case of having a screen, and interrupting the conversation flow may make the user feel unsmooth);
  • the robot According to the content of the target jump topic, the robot generates a question to the user, and determines whether to jump according to the user's answer (no screen is required). For example, ask the user "Do you want to know other products?" "Do you want to know about ** products?" If the user's answer is yes, the dialog jumps to the target jump topic. If the user negates, reselect the target hop. Turn the topic.
  • the robot using the robot dialogue method of the present invention has a clear dialogue intention (ie, a target topic). Through a series of pre-customized topics and a plan for the topic jump, the robot will ask the user according to the current conversation topic and guide the user. Continue to talk about the next topic. In this way, the robot can inspire users to learn more related topics through continuous questioning and guidance, and finally achieve the intended communication intention (ie, reach the target topic).
  • the present invention can help an enterprise's business personnel to construct various intent-oriented service robots, such as sales assistants, marketing assistants, recruitment assistants, tour guide assistants, and the like.
  • a good salesperson will prepare a series of topics (company introduction, product introduction, case introduction, plan and quotation) in advance. No matter where the customer talks about the topic, I hope to introduce these aspects into the information. At the same time, it is finally going to fall on the discussion of the plan and the quotation.
  • the present invention is to give such capabilities to robots.
  • the existing dialogue management system also mainly tracks the user's intention and plans the dialogue process based on the judgment or guess of the user's intention.
  • Such technology is developed around people (user intentions), and in a person-centered scene, such as smart speakers, smart car systems, smart TVs, etc., can be carried out relatively smoothly.
  • the embodiments of the present invention can be used to provide a series of tools to help industry experts (sales experts, education experts, medical experts, etc.) to construct robots with clear communication goals, assist experts in business training or preliminary communication with customers, thereby greatly saving these experts. Time and effort to achieve cost savings and efficiency.
  • embodiments of the present invention provide a non-transitory computer readable storage medium having stored therein one or more programs including execution instructions, the execution instructions being capable of being However, it is not limited to a computer, a server, or a network device, etc., and is read and executed for performing the above-described human-machine dialog method of the present invention.
  • embodiments of the present invention also provide a computer program product comprising a computer program stored on a non-transitory computer readable storage medium, the computer program comprising program instructions When the program instructions are executed by the computer, the computer is caused to execute any of the above-described human-machine dialog methods.
  • an embodiment of the present invention further provides an electronic device, including: at least one processor, and a memory communicatively coupled to the at least one processor, wherein the memory is stored with the at least An instruction executed by a processor, the instruction being executed by the at least one processor to enable the at least one processor to perform a human-machine dialog method.
  • embodiments of the present invention further provide a storage medium having stored thereon a computer program, wherein the program is executed by the processor in a human-machine dialog method.
  • FIG. 5 is a schematic diagram of a hardware structure of an electronic device for performing a human-machine dialog method according to another embodiment of the present application. As shown in FIG. 5, the device includes:
  • processors 510 and memory 520 one processor 510 is taken as an example in FIG.
  • the apparatus for executing the human-machine dialog method may further include: an input device 530 and an output device 540.
  • the processor 510, the memory 520, the input device 530, and the output device 540 may be connected by a bus or other means, as exemplified by a bus connection in FIG.
  • the memory 520 is a non-volatile computer readable storage medium, and can be used for storing a non-volatile software program, a non-volatile computer executable program, and a module, such as a program corresponding to the human-machine dialog method in the embodiment of the present application. Instruction/module.
  • the processor 510 executes various functional applications and data processing of the server by executing non-volatile software programs, instructions, and modules stored in the memory 520, that is, implementing the human-machine dialog method of the above method embodiment.
  • the memory 520 may include a storage program area and an storage data area, wherein the storage program area may store an operating system, an application required for at least one function; the storage data area may store data created according to usage of the human-machine dialog device, and the like. Further, the memory 520 may include a high speed random access memory, and may also include a nonvolatile memory such as at least one magnetic disk storage device, flash memory device, or other nonvolatile solid state storage device. In some embodiments, the memory 520 can optionally include a memory remotely located relative to the processor 510 that can be connected to the human machine dialog device over a network. Examples of such networks include, but are not limited to, the Internet, intranets, local area networks, mobile communication networks, and combinations thereof.
  • Input device 530 can receive input numeric or character information and generate signals related to user settings and function control of the human-machine dialog device.
  • the output device 540 can include a display device such as a display screen.
  • the one or more modules are stored in the memory 520, and when executed by the one or more processors 510, perform a human-machine dialog method in any of the above-described method embodiments.
  • the electronic device of the embodiment of the present application exists in various forms, including but not limited to:
  • Mobile communication devices These devices are characterized by mobile communication functions and are mainly aimed at providing voice and data communication.
  • Such terminals include: smart phones (such as iPhone), multimedia phones, functional phones, and low-end phones.
  • Ultra-mobile personal computer equipment This type of equipment belongs to the category of personal computers, has computing and processing functions, and generally has mobile Internet access.
  • Such terminals include: PDAs, MIDs, and UMPC devices, such as the iPad.
  • Portable entertainment devices These devices can display and play multimedia content. Such devices include: audio, video players (such as iPod), handheld game consoles, e-books, and smart toys and portable car navigation devices.
  • the server consists of a processor, a hard disk, a memory, a system bus, etc.
  • the server is similar to a general-purpose computer architecture, but because of the need to provide highly reliable services, processing power and stability High reliability in terms of reliability, security, scalability, and manageability.
  • the device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separate, and the components displayed as units may or may not be physical units, ie may be located A place, or it can be distributed to multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of the embodiment.

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Computational Linguistics (AREA)
  • General Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • Mathematical Physics (AREA)
  • Databases & Information Systems (AREA)
  • Human Computer Interaction (AREA)
  • Artificial Intelligence (AREA)
  • Evolutionary Computation (AREA)
  • Audiology, Speech & Language Pathology (AREA)
  • Multimedia (AREA)
  • Software Systems (AREA)
  • Acoustics & Sound (AREA)
  • Computing Systems (AREA)
  • Health & Medical Sciences (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Algebra (AREA)
  • Computational Mathematics (AREA)
  • Mathematical Analysis (AREA)
  • Mathematical Optimization (AREA)
  • Probability & Statistics with Applications (AREA)
  • Pure & Applied Mathematics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Animal Behavior & Ethology (AREA)
  • Machine Translation (AREA)
  • Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

本发明公开一种人机对话方法，包括：围绕目标话题确定设定数量的跳转话题，并基于设定数量的跳转话题之间的关联强度生成收敛至目标话题的话题跳转图谱，其中，每一个跳转话题至其能够跳转到的下游层级的k个跳转话题的k个路径片段配置有初始的跳转概率；在对用户的对话请求进行初始应答之后，从话题跳转图谱中选择用户的对话请求所属的跳转话题作为初始话题，以用于向用户进行第一轮推荐；当完成初始话题的人机对话之后，根据从所述初始话题跳出至下游层级的k个跳转话题的跳转概率确定将要跳转到的跳转话题；采用逐级推荐的方式从初始话题逐步引导至所述目标话题。本发明实现了基于明确沟通目标的人机对话，并且沟通更顺畅，效率更高。

Description

人机对话方法及电子设备 技术领域
本发明涉及人工智能技术领域，尤其涉及一种人机对话方法及电子设备。
背景技术
目前市场上各种对话机器人纷纷涌现，有的以个人助理形态出现(siri，cortana，灵犀等)，有的以聊天机器人出现(小冰、度秘)，还有的则内置在智能音箱、智能车载设备、智能电视等终端内。如果分析这些机器人背后的对话技术，大体可分为四种类型：
任务型对话：针对用户常常需要的任务而专门定制的垂直领域对话：例如：订餐，订票，寻找音乐、电影或某种商品等等。用户讲一句话后，首先会判断是哪个任务的需求，并提取用户的需求参数(如：出发地，餐馆类型等)。如果事先定义的必要参数没有收集全，机器会通过提问来获取信息。因此，任务型对话通常是多轮对话。用户也可能在对话过程中不断修改或完善自己的需求。
基于问答对的对话：知识以问答对的形式组织，将用户的提问跟问答对中的问句比较，找到最接近的，并将答案返回。这种对话常用于客户服务机器人和闲聊机器人。这类问答服务多数是单论的，有些具备一些多轮对话能力，主要涉及简单的上下文处理和指代消歧。
基于知识图谱的对话：用户用自然语言/语音来查询以三元组形式存储的事实性知识。比如“姚明的女儿有多高？”。在进行这类对话的时候，机器人需要一定的推理能力，上面这句话实际上由两步完成(1)姚明的女儿->姚沁蕾；(2)姚沁蕾的身高->160cm。这类对话大多数是单轮的，有些具备一定的多轮对话能力，主要是指代消歧(她几岁了？她->姚沁蕾)。
生成式聊天：通过训练神经网络模型，可以根据一个用户问题自动生 成一个答复。这类聊天没有明确的沟通目标和领域限定，用户讲一句话，系统就自动生成一句回复，回复和问题有一定关联，但没有明确的沟通目标。也有人称这类对话为开放域聊天。开放域聊天在现有的人机对话系统中，主要起到拉近距离，建立信任关系，情感陪伴，顺滑对话过程(例如，在任务类对话无法满足用户需求时)和提高用户粘性的作用。
现有技术中的各种智能机器人，有些采用其中一种形式，有些则是几种形式的组合体。在所有这些对话形式中，基本上都是人是主动方，机器是被动方，机器等待人来提问，然后给出答案。在任务型的对话中，机器也会向人提问的，但是在非常明确定义的情况下。比如买机票任务，事先定义三个必需参数：出发地点、到达地点、出发时间。只有用户告知全了，机器人才能向机票服务系统发出查询，如果不全，机器人就会通过提问来获取缺失的信息。但整体上，机器还是被动等待状态。
综上所述，目前各种问答机器人都是被动型的，机器人处于等待用户发问状态，用户主动问问题后，机器人开始理解用户意图，并根据用户意图来查询和反馈答案，或者执行动作。虽然有多轮对话，但也主要是反问用户一些必要参数(哪天出发？你喜欢什么颜色的？)。机器人本身没有明确的对话意图。在这样的对话系统中，如果用户想不起来话题或者具体的提问方式，对话也就进行不下去了，不能做到想用户之所想，为用户提供更好的交互体验。
发明内容
本发明实施例提供一种人机对话方法及电子设备，用于至少解决上述技术问题之一。
第一方面，本发明实施例提供一种人机对话方法，应用于电子设备，所述方法包括：
围绕目标话题确定设定数量的跳转话题，并基于所述设定数量的跳转话题之间的关联强度生成收敛至所述目标话题的话题跳转图谱，其中，每一个跳转话题至其能够跳转到的下游层级的k个跳转话题的k个路径片段配置有初始的跳转概率；
在对用户的对话请求进行初始应答之后，从所述话题跳转图谱中选择 所述用户的对话请求所属的跳转话题作为初始话题，以用于向用户进行第一轮推荐；
当完成所述初始话题的人机对话之后，根据从所述初始话题跳出至下游层级的k个跳转话题的跳转概率确定将要跳转到的跳转话题，以用于向用户进行下一轮推荐；
采用逐级推荐的方式从所述初始话题逐步引导至所述目标话题。
第二方面，本发明实施例提供另一种人机对话方法，应用于电子设备，所述方法包括：
确定与指定领域相关的设定数量的跳转话题，并基于所述设定数量的跳转话题之间的关联强度构建话题之间的话题跳转图谱，其中，每一个跳转话题至其能够跳转到的下游层级的k个跳转话题的k个路径片段配置有初始的跳转概率；
在对用户的对话请求进行初始应答之后，确定所述初始应答所属的跳转话题为初始话题，以用于向用户进行第一轮推荐；
当完成所述初始话题的人机对话之后，根据从所述初始话题跳出至下游层级的k个跳转话题的跳转概率确定将要跳转到的跳转话题，以围绕所述将要跳转到的跳转话题进行下一轮推荐；
采用逐轮推荐的方式继续推荐所述话题跳转图谱中的跳转话题以进行人机对话。
第三方面，本发明实施例提供一种非易失性计算机可读存储介质，所述存储介质中存储有一个或多个包括执行指令的程序，所述执行指令能够被电子设备(包括但不限于计算机，服务器，或者网络设备等)读取并执行，以用于执行本发明上述任一项人机对话方法。
第四方面，提供一种电子设备，其包括：至少一个处理器，以及与所述至少一个处理器通信连接的存储器，其中，所述存储器存储有可被所述至少一个处理器执行的指令，所述指令被所述至少一个处理器执行，以使所述至少一个处理器能够执行本发明上述任一项人机对话方法。
第五方面，本发明实施例还提供一种计算机程序产品，所述计算机程序产品包括存储在非易失性计算机可读存储介质上的计算机程序，所述计算机程序包括程序指令，当所述程序指令被计算机执行时，使所述计算机 执行上述任一项人机对话方法。
本发明实施例的有益效果在于：通过预先基于跳转话题之间的关联强度所建立的话题跳转图谱，并在人机对话过程中以确定对话人的首次对话请求为所对应的话题跳转图普中的跳转话题，之后再按照跳转话题之间的关联强度主动为用户推荐与用户当前所谈论话题相关的下一个跳转话题，使得计算机无论是回答用户主动提出的问题，还是基于用户提问向用户进行相关话题的推荐，都能够始终处于高效运转状态，提升了人机对话过程中计算机的利用率；此外，还保证了人与机器之间对话的顺利进行，提升了用户在进行人机对话中的体验。
附图说明
为了更清楚地说明本发明实施例的技术方案，下面将对实施例描述中所需要使用的附图作一简单地介绍，显而易见地，下面描述中的附图是本发明的一些实施例，对于本领域普通技术人员来讲，在不付出创造性劳动的前提下，还可以根据这些附图获得其他的附图。
图1为本发明的人机对话方法的一实施例的流程图；
图2为本发明的人机对话方法的另一实施例的流程图；
图3为本发明的人机对话方法的又一实施例的流程图；
图4为本发明的人机对话方法的再一实施例的流程图；
图5为本发明的电子设备的一实施例的结构示意图。
具体实施方式
为使本发明实施例的目的、技术方案和优点更加清楚，下面将结合本发明实施例中的附图，对本发明实施例中的技术方案进行清楚、完整地描述，显然，所描述的实施例是本发明一部分实施例，而不是全部的实施例。基于本发明中的实施例，本领域普通技术人员在没有作出创造性劳动前提下所获得的所有其他实施例，都属于本发明保护的范围。
需要说明的是，在不冲突的情况下，本申请中的实施例及实施例中的特征可以相互组合。
本发明可以在由计算机执行的计算机可执行指令的一般上下文中描 述，例如程序模块。一般地，程序模块包括执行特定任务或实现特定抽象数据类型的例程、程序、对象、元件、数据结构等等。也可以在分布式计算环境中实践本发明，在这些分布式计算环境中，由通过通信网络而被连接的远程处理设备来执行任务。在分布式计算环境中，程序模块可以位于包括存储设备在内的本地和远程计算机存储介质中。
在本发明中，“模块”、“装置”、“系统”等等指应用于计算机的相关实体，如硬件、硬件和软件的组合、软件或执行中的软件等。详细地说，例如，元件可以、但不限于是运行于处理器的过程、处理器、对象、可执行元件、执行线程、程序和/或计算机。还有，运行于服务器上的应用程序或脚本程序、服务器都可以是元件。一个或多个元件可在执行的过程和/或线程中，并且元件可以在一台计算机上本地化和/或分布在两台或多台计算机之间，并可以由各种计算机可读介质运行。元件还可以根据具有一个或多个数据包的信号，例如，来自一个与本地系统、分布式系统中另一元件交互的，和/或在因特网的网络通过信号与其它系统交互的数据的信号通过本地和/或远程过程来进行通信。
最后，还需要说明的是，在本文中，诸如第一和第二等之类的关系术语仅仅用来将一个实体或者操作与另一个实体或操作区分开来，而不一定要求或者暗示这些实体或操作之间存在任何这种实际的关系或者顺序。而且，术语“包括”、“包含”，不仅包括那些要素，而且还包括没有明确列出的其他要素，或者是还包括为这种过程、方法、物品或者设备所固有的要素。在没有更多限制的情况下，由语句“包括……”限定的要素，并不排除在包括所述要素的过程、方法、物品或者设备中还存在另外的相同要素。
如图1所示，本发明的实施例提供一种用于机器人对话的话题跳转图谱优化方法，包括：
S11、围绕目标话题确定设定数量的跳转话题，并基于所述设定数量的跳转话题之间的关联强度生成收敛至所述目标话题的话题跳转图谱，其中，每一个话题至其能够跳转到的下游层级的k个话题的k个路径片段配置有初始的跳转概率。其中，k个初始跳转概率可以表示为P 1-P k。
本发明实施例中的目标话题可以是预先确定的至少一个话题，可以采 用一系列(例如50-100个)跟目标话题相关话题来生成话题跳转图谱，可以对确定的设定数量的跳转话题进行分类，例如，分类成：关键话题(其中有部分话题可能对达成目标非常重要，尽量要让用户了解)、相关话题(对关键话题的一些补充，有助于达成沟通目标)和趣味话题(增加沟通趣味性的话题)。每一个跳转话题和/或目标话题中分别包含多个知识点，所述知识点采用问答对表示和/或采用知识图谱表示。
跳转话题之间的关联强度根据跳转话题的类推关系、引用关系等多种策略确定(得到一系列话题后根据话题间的关联关系构造话题跳转图谱。例如，为两个在概念上存在父子、包含、并列、引用或类推关系的话题间建立跳转关系；也可以根据话题包含的两组知识点的语义相似度来设定跳转关系；还可以可根据相关话题在文献、书籍、网络信息中相关关联的强度来建立跳转关系，本发明对此不作限定)，为每一个跳转话题建立下一步可跳转的话题路径，并为每个跳转路径分配一个初始概率(每一条路径的初始概率可以相同也可不同，例如，不同类型的关系对应不同的初始概率，或者根据语义相似度的大小、话题之间的关联强度等指标计算出对应的初始概率)。如果行业专家有特殊需求，也可以指定跳转路径和概率。经过话题规划，就得到一个话题之间的有限状态机，每个话题代表一个状态，每个话题跳转是一条带概率的有向边。为了保证对话过程有一定的多样性，同时也能有效达成对话目标，通常为一个话题设定跳转目标在总话题数量的10％左右(实际应用时根据总的话题量适当调整)。
需要说明的是本发明实施例中所述的下游层级，仅仅是相对于当前跳转话题而言的，在整个话题跳转图谱中并不存在上下游层级这种关系，例如对于当前话题A1，如果话题A1能够跳转到的话题包括话题B、C、D，那么这时话题B、C、D就是话题A1的下游层级的跳转话题，如果还存在话题A和话题A2，并且话题A1、A2分别是话题A的下游层级，但是话题A2不能够跳转至话题B、C、D(即，与这些话题之间的不存在跳转路径)。所以，即便是相对于话题A1、A2同属于话题A的下游层级，并且话题B、C、D又是话题A1的下游层级，由于话题A2不能够跳转至话题B、C、D所以相互之间不存在上下游层级的关系。
S12、在对用户的对话请求进行初始应答之后，从所述话题跳转图谱 中选择所述用户的对话请求所属的话题作为初始话题，以用于向用户进行第一轮推荐。
例如，用户提问“你们公司有哪些产品？”，然后对此进行自然语言理解，并返回答复“XXX公司专注于垂直领域下的智能对话式交互，主要产品包括智能车载方案、智能家居方案、智能机器人等物联网及泛物联网领域。”。确定包含用户提问的话题为初始话题，具体地可以是包含有与用户提问相同或者相近似的语句(包含在对话对中)，或者知识图谱中包含有与用户提问相关联的知识点。
S13、当完成所述初始话题的人机对话之后，根据从所述初始话题跳出至下游层级的k个跳转话题的跳转概率确定将要跳转到的跳转话题，以用于向用户进行下一轮推荐。
具体地，基于话题跳转图谱确定自初始话题可跳转的下游的跳转话题，并根据跳转概率从下游跳转话题中选择目标跳转话题，并从确定的该目标跳转话题中所包含的知识点中选择问题推荐给用户进行人机对话，当完成当前目标跳转话题的对话之后，再进一步确定当前目标跳转话题可跳转至的下游跳转节点……依次类推，直到确定出初始跳转话题到目标话题的整条跳转路径。
具体地，根据跳转概率从下游跳转话题中选择目标跳转话题可以为按照概率采样来选择目标话题。例如，话题A有三条跳出路径分别跳转至话题B、C、D，并且分别对应的跳转概率为0.5,0.3,0.2，那么按概率采样意味着，如果进行100次人机对话交互，50次走第一条路径(A-B)，30次走第二条(A-C)，20次走第三条(A-D)。
S14、采用逐级推荐的方式从所述初始话题逐步引导至所述目标话题。
本发明实施例的机器人对话方法中的话题跳转图谱是基于围绕目标话题的多个跳转话题规划得到的，并且该跳转图谱最终跳转收敛至目标话题，从而使得本发明实施例的机器人对话能够具有一定的对话意图(目标话题)，并且能够通过向着目标话题跳转收敛的外围跳转话题逐步将最终话题跳转至对话意图的目标话题，从而有助于用户与机器人之间对话的顺利进行，且不至于将话题聊飞(偏离主题，即，偏离目标话题)。此外，由于采用逐级推荐下游跳转话题的方式，所以能够使得机器人在对话过程 中在一定程度上具备主动性，这样一方面可以推进对话的顺利进行并最终跳转至目标话题，完成对话目的，另一方面在推荐过程中由于是逐级推荐而并非直接生硬的进入目标话题，不会引起用户的反感，还提升了用户体验。
本发明实施例中的人机对话方法可以体现为多种产品形态，例如，“销售助理”、“企业名片”等。其中，对于销售助理，可以以达成销售目标为目的，所以可以确定“询价”、“订单”、“售后”之类的话题为目标话题，确定围绕具体要销售的产品的周边话题为跳转话题，以基于跳转话题之间的关联强度生成收敛至目标话题的话题跳转图谱，从而用于进行以销售为目的的人机对话。“企业名片”可以定位成以传播企业关键信息或者达成合作为目标功能，这时类似于“销售助理”确定目标话题以及相关的跳转话题来构建话题跳转图谱以进行人机对话。因此，本发明的人机对话方法能够在为用户提供友好的交互体验的过程中促成企业的宣传与产品销售。
示例性地，销售助理和企业名片等可以表现为应用程序或者微信小程序等，本发明对此不作限定。以下以表现为应用程序的销售助理为例进行详细说明：
不同的公司可以针对自己的产品定制各自的销售助理。示例性地，以销售相关的话题(例如，询价、订单、售后等话题)确定为目标话题，以公司的其它相关信息所属的话题(例如，公司简介、案例、技术原理等话题)为跳转话题，从而构造出话题跳转图谱。
实际应用中，用户可以点击打开销售助理应用程序，在该应用程序的显示界面之下显示了多个导航标签(例如，公司简介、案例、技术原理、购买等)，用户可以采用语音输入的方式选择所感兴趣的导航标签，并从用户所选择的导航标签所对应的话题开始展开人机对话，并基于构建好的话题跳转图谱逐渐将话题引向目标话题。
示例性地，从跳转话题向目标话题收敛的一种跳转关系可以是：公司简介→案例→技术原理→购买。并给出以下一种人机对话的示例：
用户：公司简介
销售助理：XXX公司成立于X年，专注于垂直领域下的智能对话式 交互，主要产品包括智能车载方案、智能家居方案、智能机器人等物联网及泛物联网领域。
销售助理：你是否想要了解公司相关案例？
用户：是的。
销售助理：本公司语音助理产品已经应用于XX智能音箱，麦克风阵列产品已经应用于XX汽车的车载音箱等。
用户：麦克风阵列还有哪些应用案例。
销售助理：本公司的双麦阵列还应用在了……。
销售助理：你是否想要了解双麦阵列的原理？
用户：是的。
销售助理：双麦阵列的原理是……。
销售助理：你是否考虑使用我们公司的双麦阵列产品？
用户：……。
由以上示例可以看出，整个人机对话过程从用户所选择话题出发，逐步的在回答用户与并基于话题跳转图谱推荐反问用户的方式，将话题引向了目标话题(销售话题)。
如图2所示，在一些实施例中，所述人机对话方法还包括：
S21、当采用逐级推荐的方式跳转至所述目标话题并完成人机对话后，确定所述初始话题至所述目标话题的跳转路径。
本发明实施例中所述完成人机对话指的是用户对本次人机对话满意，可以在结束人机对话之前通过选择评价的方式或者回答机器人主动提出是否满意本次对话的问题的方式类确定，如果用户表示满意本次人机对话，则判定为本次人机对话完成，并执行后续步骤。
S22、上调所述跳转路径上的各个路径片段的跳转概率，更新所述话题跳转图谱，以用于下一次人机对话。
由于在步骤S21中已经确定圆满的完成了本次人机对话，这就表明本次人机对话过程中向用户所推荐并引导至目标话题的路径所经过的跳转话题都是用户所比较乐于谈及的话题，所以为了在下一次人机对话中能够更高概率的推荐被用户所认可的跳转话题，所以在本发明的实施例中对这 些指向目标话题并被用户所认可的跳转话题设置更大的跳转概率。
本发明实施中在进行人机对话的同时，还实现了话题跳转图谱的不断的优化，从而能够再下一次进行人机对话时能够具有更优的性能，响应速度更快更准确，提升用户体验。
此外，经过跳转概率的调整，可以在后续再次进行人机对话时减少计算机的计算量。例如，可以使得计算机原本需要计算10步跳转概率才能够达到目标话题的情况，优化计算机只需要计算5步跳转概率就可以达到目标话题，减少了计算机的计算量，提高了达到目标话题的效率。
在一些实施例中，每一个跳转话题至其能够跳转到的下游层级的k个跳转话题的k个路径片段初始的跳转概率之和为1；
在上调所述跳转路径上的第i个跳转话题至下游层级的第i+1个跳转话题之间的路径片段P(i,i+1)的跳转概率的同时，相应的降低第i个跳转话题至下游层级的除所述第i+1个跳转话题之外的其它k-1个跳转话题的k-1个路径片段的跳转概率，以保持所述第i个跳转话题至下游层级的k个路径片段的跳转概率之和仍然为1。其中，第一跳转概率可以采用P f1表示，n-1个第二跳转概率可以采用P’ f1-P’ f(n-1)表示。
本实施例中，同一个跳转话题与其下游层级的k个跳转话题之间的路径片段的跳转概率之和遵循归一化原则，能够保持整个话题跳转图谱的稳定性，并且已于话题跳转图谱的优化管理工作，以更好的进行人机对话。
如图3所示，本发明实施例还提供另一种人机对话方法，包括：
S31、确定与指定领域相关的设定数量的跳转话题，并基于所述设定数量的跳转话题之间的关联强度构建话题之间的话题跳转图谱，其中，每一个跳转话题至其能够跳转到的下游层级的k个跳转话题的k个路径片段配置有初始的跳转概率；本发明实施例中的指定领域可以是教学、培训、市场宣传等领域。
S32、在对用户的对话请求进行初始应答之后，确定所述初始应答所属的跳转话题为初始话题，以用于向用户进行第一轮推荐；
S33、当完成所述初始话题的人机对话之后，根据从所述初始话题跳出至下游层级的k个跳转话题的跳转概率确定将要跳转到的跳转话题，以 围绕所述将要跳转到的跳转话题进行下一轮推荐；
S34、采用逐轮推荐的方式继续推荐所述话题跳转图谱中的跳转话题以进行人机对话。
通过预先基于跳转话题之间的关联强度所建立的话题跳转图谱，并在人机对话过程中以确定对话人的首次对话请求为所对应的话题跳转图普中的跳转话题，之后再按照跳转话题之间的关联强度主动为用户推荐与用户当前所谈论话题相关的下一个跳转话题，从而保证了人与机器之间对话的顺利进行，提升了用户在进行人机对话中的体验。本发明实施例的机器人对话方法中，由于在人机对话过程中始终能够推荐与用户当前谈论话题相关联的另一个话题(即，被推荐话题)，从而能够保证人机对话过程中所谈论的话题始终是用户所感兴趣的话题，以至于能够让用户在不知不觉中与机器人之间谈论诸多话题，获取多方面多维度的知识。
本发明实施例中的人机对话方法可以体现为多种产品形态，例如，“企业宣传名片”、“故事机”、“知识百科”等。以“企业宣传名片”为例，可以围绕企业的多个维度(例如，企业发展历程、企业文化、企业架构组成、企业优势、企业产品、企业人才、企业案例等)确定与企业相关的多个话题，以用于构建本发明实施例中的话题跳转图谱，从而基于该图谱实现能够主动推荐话题的人机对话模式，能够让用户在友好的人机对话过程中尽可能全面的了解本企业，起到企业品牌宣传的目的。
示例性地，企业宣传名片可以表现为应用程序或者微信小程序等，本发明对此不作限定。以下以表现为应用程序的企业宣传名片为例进行详细说明：
不同的公司可以针对本公司的组织架构定制各自的企业宣传名片机器人。示例性地，以公司简介、案例、技术原理、产品、产品购买等话题之间的关联性构造出话题跳转图谱。
实际应用中，用户可以点击打开企业宣传名片程序，在该应用程序的显示界面之下显示了多个导航标签(例如，公司简介、案例、技术原理、产品、产品购买等)，用户可以采用语音输入的方式选择所感兴趣的导航标签，并从用户所选择的导航标签所对应的话题开始展开人机对话，并基于构建好的话题跳转图谱向用户传播公司信息。
示例性地，从话题跳转图谱中的话题之间的跳转关系可以是：公司简介→产品→技术原理→案例→产品→技术原理→案例→产品购买→产品……。并给出以下一种人机对话的示例：
用户：公司简介
企业宣传名片：XXX公司成立于X年，专注于垂直领域下的智能对话式交互，主要产品包括智能车载方案、智能家居方案、智能机器人等物联网及泛物联网领域。
企业宣传名片：你是否想要了解公司相关产品？
用户：是的。
企业宣传名片：本公司产品有：智能车载解决方案、智能家居解决方案、智能机器人解决方案和语音输入板。
企业宣传名片：你是否想要了解公司相关产品的技术原理？
用户：是的，想了解智能车载解决方案。
企业宣传名片：智能车载解决方案的技术原理是……。
企业宣传名片：你还想了解一下相关案例吗？
用户：是的
企业宣传名片：本公司的智能车载解决方案已经成功应用于XX汽车之上。
企业宣传名片：你是否还想了解我们公司的智能家居解决方案？
用户：……。
本实施例中，在人机对话过程中，通过不同话题构成的话题跳转图谱，能够将与当前企业相关的各种企业信息传播给用户。并且由于话题跳转图谱中的各个话题之间的跳转连接关系是根据各话题之间的关联性确定的，使得每一次推荐给用户的话题都是与用户感兴趣的话题，从而使得计算机无论是回答用户主动提出的问题，还是基于用户提问向用户进行相关话题的推荐，都能够始终处于高效运转状态，提升了人机对话过程中计算机的利用率。
如图4所示，在一些实施例中，每一个跳转话题至其能够跳转到的下 游层级的k个跳转话题的k个路径片段初始的跳转概率之和为1；在进行跳转话题推荐的过程中还包括：
S41、当选择第一跳转话题的下游层级的其中一个第二跳转话题进行下一轮推荐时；
S42、若用户接受所述其中一个第二跳转话题，则增大所述第一跳转话题至所述其中一个第二跳转话题的路径片段的跳转概率，减小所述第一跳转话题至下游层级的其它第二跳转话题的路径片段的跳转概率，以保持所述第一跳转话题至下游层级的所有第二跳转话题的路径片段的跳转概率之和为1；
S43、若用户不接受所述第二跳转话题，则减小所述第一跳转话题至所述其中一个第二跳转话题的路径片段的跳转概率，增大所述第一跳转话题至下游层级的其它第二跳转话题的路径片段的跳转概率，以保持所述第一跳转话题至下游层级的所有第二跳转话题的路径片段的跳转概率之和为1，并更换选择所述第一跳转话题的下游层级中的另外一个第二跳转话题进行下一轮推荐。
本发明实施例中根据每一次为用户推荐跳转话题之后用户的反应情况来相应的修正至被推荐跳转话题的跳转概率，从而能够使得话题跳转图谱得到及时的优化，以为随后进行的人机对话服务提供更友好的对话保障。
此外，经过跳转概率的调整，可以在后续再次进行人机对话时减少计算机的计算量。例如，企业名片对话机器人，想要传播企业关键信息，采用优化之后的话题跳转图谱，能够使得计算机每一次推荐给用户的跳转话题都是用户感兴趣的话题(不存在或者至少减少了计算机推荐了用户不喜欢的话题，而需要计算机重新计算并给用户推荐新的话题的情况)，从而能够计算机能够节省大量重复计算重复推荐的工作量，保证了计算机处理能力的高效利用。例如，可以使得计算机原本需要计算10步跳转才能够达到传播企业关键信息的目的，优化计算机只需要计算5步跳转就可以达到传播企业关键信息的目的，减少了计算机的计算量，提高了达到目标话题的效率。
在一些实施例中，还包括：
当累计更换选择的次数超过预定次数时，重新选择与所述第一跳转话题独立的话题。
本发明实施例中的预定次数可以是3-5次，将给用户推荐跳转话题的次数限制在预定次数之内，避免了过多的推荐用户不接受的话题内容给用户造成的骚扰，并且为跳转至被用户所拒绝的跳转话题的跳转配置小于初始跳转概率的跳转概率，从而再以后的应用于优化中，将曾经被用户拒绝的跳转话题再次被推荐的概率降低，避免再次给用户造成困扰，从而提升了人机对话的友好性，进一步提升用户体验。
在一些实施例中，在话题跳转图谱的动态调整中，根据不同策略自动为每个话题挑选若干跳转话题并设置跳转概率。但是这些跳转未必会被用户接受。因此在实际使用过程中还需要根据用户使用情况不断调整这些跳转路径和跳转概率。
跳转概率的在线学习：
在一些实施例中，如果用户接受了跳转推荐(或点击机器的跳转建议表示肯定)，则该跳转路经m获得正向reward，相应的对该路径上的跳转概率增加x 1(P m＝P m+x 1)，P m为该路径的跳转概率，同时从该话题跳出至下游层级的其他路径片段的跳转概率分别降低x 1/(N-1)，即P n＝P n-x 1/(N-1)，以保证一个话题所有跳出路径的概率之和始终为1。N为跳出路径的个数(即，当前下游层级中跳转话题的个数)，n＝1，...,N-1，且n≠m，x 1为事先设定的一个0～1之间的值，通常是一个很小的正数，比如x 1＝0.001；如果用户明确拒绝跳转建议，则该跳转路径m获得负向reward，相应的该路径上的跳转概率降低y 1(P m＝P m-y 1)，同时从该话题跳出的其它路径的跳转概率分别增加y 1/(N-1)，即P m＝P m+y 1/(N-1)，N为跳出路径的个数，n＝1，...,N，且n不等于m，y 1为事先设定的一个0～1之间的值，通常是一个很小的正数，比如y 1＝0.0015；类似的，如果用户没有点击推荐话题，该跳转路径获的跳转概率下降y 2，其它路径的跳转概率增加y 2/(N-1)，y 2通常是一个很小的正数，比如y 2＝0.0005；x 1,y 1,y 2的设置也可以根据跳出话题的使用量来动态调整。使用次数越多，这几个取值应该越小。这里描述的是一个根据每次使用情况实时调整跳转概率的方 法。实际应用中也可以变通成以每小时、每天的使用情况的进行定时调整算法。
跳转路径的优化：当某条跳转路径的概率小于一个预先设定的阈值的时候，则将该路径取消。同时从还没有尝试过的路径中选择一条新的路劲添加上来，并赋予一个初始跳转概率z 1(其它路径的概率需要适当调整保证所有路径上的概率之和为1)。
在一些实施例中，在多次人机对话之后，若存在至少一个路径片段的跳转概率低于预定阈值，则断开所述话题跳转图谱中的所述至少一个路径片段，更新所述话题跳转图谱，以用于下一次人机对话。
本发明实施例中，在人机对话过程中实现了对话题跳转图谱的优化，如果跳转话题至另一跳转话题的路径片段的跳转概率值小于预定阈值时(这样的路径片段存在至少一个)，直接将该某一跳转话题删除，以免低概率跳转路径对以后人机对话造成干扰，另一方面也可以起到简化话题跳转图谱，从而加快人机对话实时性的效果(因为，当整个话题跳转图谱足够大时，实际所存在的跳转话题以及跳转话题之间的路径片段是非常巨大的，这时将小概率的路径片段断开，在很大程度上简化了话题跳转图谱的结构，相应的也极大的提升了基于该话题跳转图谱的人机对话过程的实时性)。
在一些实施例中，在断开所述话题跳转图谱中的所述至少一个路径片段时，为所述至少一个路径片段的端点的跳转话题配置一个新的跳转话题。本发明实施例中，在构建话题跳转图谱的过程中，为每一个当前话题选定的多个可以跳转到的跳转话题是从与当前话题相关联的所有话题中选择出来的，未被选中的相关联的话题存储为相应的当前话题的候选跳转话题集，本实施例中，为所述至少一个路径片段的端点的跳转话题配置的新的跳转话题选自该路径片段的另一个端点的话题的候选跳转话题集。
当确定的与当前话题相关联的所有话题全部设定为该当前话题的跳转话题，则在本实施例中，为所述至少一个路径片段的端点的跳转话题所配置的新的跳转话题可以是随机选自话题跳转图谱的跳转话题，并赋予跳 转概率。
本发明实施例中，在删除了跳转概率低于预定阈值的至少一个路径片段中的下游跳转话题的同事还引进了新的跳转话题，这样既保证了对原本不合理的跳转图谱实现了优化(将被用户所频繁拒绝的跳转话题删除)，而且及保证了为用户推荐跳转话题的多样性(保证当前跳转话题能够跳转到的目标跳转话题的数量不至于过少)，又有可能将初始构建跳转图谱时错误判断的跳转节点重新找回来，以构成更加合理高效的跳转图谱。
在一些实施例中，还包括：按照用户属性对用户进行分类，以按照用户的类别生成相对应的话题跳转图谱。
本实施了中通过按照用户属性分类之后，针对不同类别的用户生成相应类别的跳转图谱，从而能够在用户使用发明实施例中的跳转图谱的机器人进行对话时可以获取更加适合与自己的机器人工作模式，从而能够使得人机对话过程更加的顺畅，提升用户体验。
因为，发明人在实现本发明的过程中发现，当用户的身份不同时，其感兴趣的话题是不一样的。例如，对于销售助理类的机器人，其需要向用户提供公司各种相关信息的介绍。但是，当咨询的用户是采购人员时，其所关注的问题可能更多的集中在产品的性价比方面，从而针对这类人群所构建优化得到的跳转图谱必然是以产品的性价比方面的话题为主的；当咨询的用户是技术研发人员时，其所关注的问题可能更多集中在关于产品的性能原理等方面话题的介绍，从而针对此类人群所构建得到的跳转图谱必然是以产品性能原理方面的话题为主的。
本发明实施例中，用户属性可以包括用户工作职务、用户性别、用户年龄等，可以在用户开始咨询之前要求用户填写上述信息的方式获取，也可以是其它任何方式，本发明对此不作限定。
本发明实施例中，对话的初始状态根据用户行为来确定。如果用户直接提问，则对提问进行语义理解，并找到相关的知识点，给出准确的答复。在此之后：
获取知识点所属话题以及该话题可以跳转的后续话题列表及跳转概 率。根据跳转概率进行采样，选定本次对话的目标跳转话题(被选中的跳转话题)。
本发明实施例中的话题的跳转可以有两种引导方式：
话题推荐模式：将一条或多条目标跳转话题直接推荐给用户，由用户通过点击选择(这种模式只适用于有屏幕的情况，且打断对话流，会可能让用户感觉不顺畅)；
根据目标跳转话题的内容，生成机器人对用户的提问，并根据用户回答来确定是否跳转(不需要屏幕)。比如问用户“你还想了解其它产品吗？”“你想了解**产品吗？)。如果用户的回答是肯定的，对话跳转到目标跳转话题。如果用户否定，则重新选择目标跳转话题。
采用本发明的机器人对话方法的机器人是有明确对话意图的(即，目标话题)，通过预先定制的一系列话题以及对话题跳转的规划，机器人会根据当前聊的话题向用户提问，引导用户再继续聊下一个话题。这样，机器人就能通过不断的提问和引导，启发用户了解更多的相关话题，最终达到预定的沟通意图(即，达到目标话题)。本发明可以帮助企业的业务人员构建各种意图明确的服务机器人，例如，销售助理、市场助理、招聘助理、导游助理等等。
例如，一个好的销售人员，事先会准备好一系列话题(公司介绍，产品介绍，案例介绍，方案和报价)，不管客户从哪里聊起话题，最终都希望把这几方面的信息都介绍到，同时最后要落在方案和报价的讨论上。本发明就是要赋予机器人这样的能力。
现有对话技术大多数将重心放在用户意图的理解上面，采用各种算法(规则匹配，SVM分类，深度神经网络等)解决用户意图表达的多样性问题(同一个意图可能有很多种表达方式)和意图消歧问题(一句话可能意味着多种意图)，并根据用户意图找到正确答案或执行正确的操作方面。现有的对话管理系统，也主要是追踪用户意图，并根据对用户意图的判断或猜测来规划对话流程。这样的技术是在围绕人(用户意图)展开的，在一个人为中心的场景中，比如智能音箱、智能车载系统、智能电视等，还能比较顺畅的进行。但是，如果将这些技术直接搬到企业服务场景、教育 场景等需要有明确沟通意图的情况下，发明人发现，这种被动应答式的机器人，往往不能满足需求。在很多应用场景中，需要机器人有明确的沟通目标，并能引导用户进行交互，从而达到预期目标。
本发明实施例能够用于提供一系列工具帮助行业专家(销售专家、教育专家、医疗专家等)构建有明确沟通目标的机器人，辅助专家进行业务培训或跟客户初步沟通，从而大幅节省这些专家的时间和精力，从而达到节省成本、提高效率的目的。
需要说明的是，对于前述的各方法实施例，为了简单描述，故将其都表述为一系列的动作合并，但是本领域技术人员应该知悉，本发明并不受所描述的动作顺序的限制，因为依据本发明，某些步骤可以采用其他顺序或者同时进行。其次，本领域技术人员也应该知悉，说明书中所描述的实施例均属于优选实施例，所涉及的动作并不一定是本发明所必须的。在上述实施例中，对各个实施例的描述都各有侧重，某个实施例中没有详述的部分，可以参见其他实施例的相关描述。
在一些实施例中，本发明实施例提供一种非易失性计算机可读存储介质，所述存储介质中存储有一个或多个包括执行指令的程序，所述执行指令能够被电子设备(包括但不限于计算机，服务器，或者网络设备等)读取并执行，以用于执行本发明上述任一项人机对话方法。
在一些实施例中，本发明实施例还提供一种计算机程序产品，所述计算机程序产品包括存储在非易失性计算机可读存储介质上的计算机程序，所述计算机程序包括程序指令，当所述程序指令被计算机执行时，使所述计算机执行上述任一项人机对话方法。
在一些实施例中，本发明实施例还提供一种电子设备，其包括：至少一个处理器，以及与所述至少一个处理器通信连接的存储器，其中，所述存储器存储有可被所述至少一个处理器执行的指令，所述指令被所述至少一个处理器执行，以使所述至少一个处理器能够执行人机对话方法。
在一些实施例中，本发明实施例还提供一种存储介质，其上存储有计算机程序，其特征在于，该程序被处理器执行人机对话方法。
图5是本申请另一实施例提供的执行人机对话方法的电子设备的硬件结构示意图，如图5所示，该设备包括：
一个或多个处理器510以及存储器520，图5中以一个处理器510为例。
执行人机对话方法的设备还可以包括：输入装置530和输出装置540。
处理器510、存储器520、输入装置530和输出装置540可以通过总线或者其他方式连接，图5中以通过总线连接为例。
存储器520作为一种非易失性计算机可读存储介质，可用于存储非易失性软件程序、非易失性计算机可执行程序以及模块，如本申请实施例中的人机对话方法对应的程序指令/模块。处理器510通过运行存储在存储器520中的非易失性软件程序、指令以及模块，从而执行服务器的各种功能应用以及数据处理，即实现上述方法实施例人机对话方法。
存储器520可以包括存储程序区和存储数据区，其中，存储程序区可存储操作系统、至少一个功能所需要的应用程序；存储数据区可存储根据人机对话装置的使用所创建的数据等。此外，存储器520可以包括高速随机存取存储器，还可以包括非易失性存储器，例如至少一个磁盘存储器件、闪存器件、或其他非易失性固态存储器件。在一些实施例中，存储器520可选包括相对于处理器510远程设置的存储器，这些远程存储器可以通过网络连接至人机对话装置。上述网络的实例包括但不限于互联网、企业内部网、局域网、移动通信网及其组合。
输入装置530可接收输入的数字或字符信息，以及产生与人机对话装置的用户设置以及功能控制有关的信号。输出装置540可包括显示屏等显示设备。
所述一个或者多个模块存储在所述存储器520中，当被所述一个或者多个处理器510执行时，执行上述任意方法实施例中的人机对话方法。
上述产品可执行本申请实施例所提供的方法，具备执行方法相应的功能模块和有益效果。未在本实施例中详尽描述的技术细节，可参见本申请实施例所提供的方法。
本申请实施例的电子设备以多种形式存在，包括但不限于:
(1)移动通信设备:这类设备的特点是具备移动通信功能，并且以提供话音、数据通信为主要目标。这类终端包括:智能手机(例如iPhone)、多媒体手机、功能性手机，以及低端手机等。
(2)超移动个人计算机设备:这类设备属于个人计算机的范畴，有计算和处理功能，一般也具备移动上网特性。这类终端包括:PDA、MID和UMPC设备等，例如iPad。
(3)便携式娱乐设备:这类设备可以显示和播放多媒体内容。该类设备包括:音频、视频播放器(例如iPod)，掌上游戏机，电子书，以及智能玩具和便携式车载导航设备。
(4)服务器:提供计算服务的设备，服务器的构成包括处理器、硬盘、内存、系统总线等，服务器和通用的计算机架构类似，但是由于需要提供高可靠的服务，因此在处理能力、稳定性、可靠性、安全性、可扩展性、可管理性等方面要求较高。
(5)其他具有数据交互功能的电子装置。
以上所描述的装置实施例仅仅是示意性的，其中所述作为分离部件说明的单元可以是或者也可以不是物理上分开的，作为单元显示的部件可以是或者也可以不是物理单元，即可以位于一个地方，或者也可以分布到多个网络单元上。可以根据实际的需要选择其中的部分或者全部模块来实现本实施例方案的目的。
通过以上的实施方式的描述，本领域的技术人员可以清楚地了解到各实施方式可借助软件加通用硬件平台的方式来实现，当然也可以通过硬件。基于这样的理解，上述技术方案本质上或者说对相关技术做出贡献的部分可以以软件产品的形式体现出来，该计算机软件产品可以存储在计算机可读存储介质中，如ROM/RAM、磁碟、光盘等，包括若干指令用以使得一台计算机设备(可以是个人计算机，服务器，或者网络设备等)执行各个实施例或者实施例的某些部分所述的方法。
最后应说明的是：以上实施例仅用以说明本申请的技术方案，而非对其限制；尽管参照前述实施例对本申请进行了详细的说明，本领域的普通技术人员应当理解：其依然可以对前述各实施例所记载的技术方案进行修改，或者对其中部分技术特征进行等同替换；而这些修改或者替换，并不 使相应技术方案的本质脱离本申请各实施例技术方案的精神和范围。

Claims (10)

  1. 一种人机对话方法，应用于电子设备，所述方法包括：
    围绕目标话题确定设定数量的跳转话题，并基于所述设定数量的跳转话题之间的关联强度生成收敛至所述目标话题的话题跳转图谱，其中，每一个跳转话题至其能够跳转到的下游层级的k个跳转话题的k个路径片段配置有初始的跳转概率；
    在对用户的对话请求进行初始应答之后，从所述话题跳转图谱中选择所述用户的对话请求所属的跳转话题作为初始话题，以用于向用户进行第一轮推荐；
    当完成所述初始话题的人机对话之后，根据从所述初始话题跳出至下游层级的k个跳转话题的跳转概率确定将要跳转到的跳转话题，以用于向用户进行下一轮推荐；
    采用逐级推荐的方式从所述初始话题逐步引导至所述目标话题。
  2. 根据权利要求1所述的方法，其中，还包括：
    当采用逐级推荐的方式跳转至所述目标话题并完成人机对话后，确定所述初始话题至所述目标话题的跳转路径；
    上调所述跳转路径上的各个路径片段的跳转概率，更新所述话题跳转图谱，以用于下一次人机对话。
  3. 根据权利要求2所述的方法，其中，每一个跳转话题至其能够跳转到的下游层级的k个跳转话题的k个路径片段初始的跳转概率之和为1；
    在上调所述跳转路径上的第i个跳转话题至下游层级的第i+1个跳转话题之间的路径片段P(i,i+1)的跳转概率的同时，相应的降低第i个跳转话题至下游层级的除所述第i+1个跳转话题之外的其它k-1个跳转话题的k-1个路径片段的跳转概率，以保持所述第i个跳转话题至下游层级的k个路径片段的跳转概率之和仍然为1。
  4. 一种人机对话方法，应用于电子设备，所述方法包括：
    确定与指定领域相关的设定数量的跳转话题，并基于所述设定数量的跳转话题之间的关联强度构建话题之间的话题跳转图谱，其中，每一个跳转话题至其能够跳转到的下游层级的k个跳转话题的k个路径片段配置有初始的跳转概率；
    在对用户的对话请求进行初始应答之后，确定所述初始应答所属的跳转话题为初始话题，以用于向用户进行第一轮推荐；
    当完成所述初始话题的人机对话之后，根据从所述初始话题跳出至下游层级的k个跳转话题的跳转概率确定将要跳转到的跳转话题，以围绕所述将要跳转到的跳转话题进行下一轮推荐；
    采用逐轮推荐的方式继续推荐所述话题跳转图谱中的跳转话题以进行人机对话。
  5. 根据权利要求1或4所述的方法，其中，每一个跳转话题至其能够跳转到的下游层级的k个跳转话题的k个路径片段初始的跳转概率之和为1；
    在进行跳转话题推荐的过程中还包括：
    当选择第一跳转话题的下游层级的其中一个第二跳转话题进行下一轮推荐时；
    若用户接受所述其中一个第二跳转话题，则增大所述第一跳转话题至所述其中一个第二跳转话题的路径片段的跳转概率，减小所述第一跳转话题至下游层级的其它第二跳转话题的路径片段的跳转概率，以保持所述第一跳转话题至下游层级的所有第二跳转话题的路径片段的跳转概率之和为1；
    若用户不接受所述其中一个第二跳转话题，则减小所述第一跳转话题至所述其中一个第二跳转话题的路径片段的跳转概率，增大所述第一跳转话题至下游层级的其它第二跳转话题的路径片段的跳转概率，以保持所述第一跳转话题至下游层级的所有第二跳转话题的路径片段的跳转概率之和为1，并更换选择所述第一跳转话题的下游层级中的另外一个第二跳转话题进行下一轮推荐。
  6. 根据权利要求5所述的方法，其中，还包括：
    当累计更换选择的次数超过预定次数时，重新选择与所述第一跳转话题独立的话题。
  7. 根据权利要求3或5所述的方法，其中，在多次人机对话之后，若存在至少一个路径片段的跳转概率降低至低于预定阈值，则断开所述话题跳转图谱中的所述至少一个路径片段，更新所述话题跳转图谱，以用于下一次人机对话。
  8. 根据权利要求7所述的方法，其中，在断开所述话题跳转图谱中的所述至少一个路径片段时，为所述至少一个路径片段的端点的跳转话题配置一个新的跳转话题。
  9. 根据权利要求1或4所述的方法，其中，还包括：按照用户属性对用户进行分类，以按照用户的类别生成相对应的话题跳转图谱。
  10. 一种电子设备，其包括：至少一个处理器，以及与所述至少一个处理器通信连接的存储器，其中，所述存储器存储有可被所述至少一个处理器执行的指令，所述指令被所述至少一个处理器执行，以使所述至少一个处理器能够执行权利要求1-9中任意一项所述方法的步骤。
PCT/CN2019/072603 2018-01-23 2019-01-22 人机对话方法及电子设备 WO2019144858A1 (zh)

Priority Applications (3)

Application Number Priority Date Filing Date Title
JP2020537003A JP7288446B2 (ja) 2018-01-23 2019-01-22 ヒューマンマシン対話方法および電子デバイス
US16/769,272 US11735163B2 (en) 2018-01-23 2019-01-22 Human-machine dialogue method and electronic device
EP19743677.7A EP3745283A4 (en) 2018-01-23 2019-01-22 HUMAN-MACHINE DIALOGUE AND ELECTRONIC DEVICE

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201810062270.3A CN108415932B (zh) 2018-01-23 2018-01-23 人机对话方法及电子设备
CN201810062270.3 2018-01-23

Publications (1)

Publication Number Publication Date
WO2019144858A1 true WO2019144858A1 (zh) 2019-08-01

Family

ID=63126101

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2019/072603 WO2019144858A1 (zh) 2018-01-23 2019-01-22 人机对话方法及电子设备

Country Status (5)

Country Link
US (1) US11735163B2 (zh)
EP (1) EP3745283A4 (zh)
JP (1) JP7288446B2 (zh)
CN (1) CN108415932B (zh)
WO (1) WO2019144858A1 (zh)

Families Citing this family (29)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108415932B (zh) * 2018-01-23 2023-12-22 思必驰科技股份有限公司 人机对话方法及电子设备
CN108919669B (zh) * 2018-09-11 2022-04-29 深圳和而泰数据资源与云技术有限公司 一种智能家居动态决策方法、装置和服务终端
CN110968546B (zh) * 2018-09-30 2023-04-07 北京国双科技有限公司 图谱节点跳转概率的设置方法及相关设备
CN109543019A (zh) * 2018-11-27 2019-03-29 苏州思必驰信息科技有限公司 用于车辆的对话服务方法及装置
CN109815319A (zh) * 2018-12-24 2019-05-28 联想(北京)有限公司 信息处理方法及信息处理装置
CN112017662B (zh) * 2019-05-31 2023-07-07 阿里巴巴集团控股有限公司 控制指令确定方法、装置、电子设备和存储介质
CN112035714B (zh) * 2019-06-03 2024-06-14 鲨鱼快游网络技术(北京)有限公司 一种基于角色陪伴的人机对话方法
CN110209793A (zh) * 2019-06-18 2019-09-06 佰聆数据股份有限公司 一种用于智能识别文本语义的方法
CN110516050A (zh) * 2019-07-15 2019-11-29 上海文思海辉金信软件有限公司 一种基于知识图谱的多路径训练场景的构建方法
CN110837550B (zh) * 2019-11-11 2023-01-17 中山大学 基于知识图谱的问答方法、装置、电子设备及存储介质
CN110941710B (zh) * 2019-11-27 2020-10-30 贝壳找房(北京)科技有限公司 用于实现会话的方法、装置、介质以及电子设备
CN111104504A (zh) * 2019-12-25 2020-05-05 天津中科智能识别产业技术研究院有限公司 一种基于自然语言处理和知识图谱的对话方法
CN111368046B (zh) * 2020-02-24 2021-07-16 北京百度网讯科技有限公司 人机对话方法、装置、电子设备及存储介质
CN111723189B (zh) * 2020-06-23 2021-11-16 贝壳找房(北京)科技有限公司 交互问答的提示方法和装置、存储介质、电子设备
CN112035633B (zh) * 2020-08-21 2023-07-25 腾讯科技（深圳）有限公司 数据处理方法、装置、对话设备及存储介质
CN112163087B (zh) * 2020-11-10 2023-06-23 山东比特智能科技股份有限公司 对话系统中意图冲突的解决方法、系统及装置
CN112559721B (zh) * 2020-12-25 2023-10-20 北京百度网讯科技有限公司 人机对话系统的调整方法、装置、设备、介质和程序产品
CN114691816B (zh) * 2020-12-28 2025-07-08 科沃斯商用机器人有限公司 人机多轮对话方法、装置、智能机器人和存储介质
CN112835787B (zh) * 2021-01-08 2022-09-27 展讯通信（上海）有限公司 应用程序界面跳转的路径纠正方法及装置、存储介质、终端
CN113221547B (zh) * 2021-01-21 2022-05-03 重庆邮电大学 基于信息抽取和知识图谱的试题推荐方法
CN113282708B (zh) * 2021-05-31 2023-04-07 平安国际智慧城市科技股份有限公司 机器人对话回复的方法、装置、计算机设备及存储介质
CN113268580B (zh) * 2021-07-15 2021-11-09 中国平安人寿保险股份有限公司 会话主题迁移路径挖掘方法、装置、计算机设备及介质
CN114036277B (zh) * 2021-11-15 2024-08-23 深圳壹账通智能科技有限公司 一种对话机器人路由跳转方法、装置、电子设备和介质
CN114117024B (zh) * 2022-01-27 2022-05-03 永鼎行远(南京)信息科技有限公司 一种多轮对话功能场景的平台构建方法
CN114662960A (zh) * 2022-04-02 2022-06-24 深圳市神州云海智能科技有限公司 业务流程生成方法、终端设备及计算机可读存储介质
CN115408511B (zh) * 2022-11-02 2023-01-20 深圳市人马互动科技有限公司 交互节点跳转方法及组件、对话开发系统
CN115809669B (zh) * 2022-12-30 2024-03-29 联通智网科技股份有限公司 一种对话管理方法和电子设备
US20240378306A1 (en) * 2023-05-08 2024-11-14 Nvidia Corporation Role-based large language model to enable security and accuracy
JP7461086B1 (ja) 2023-05-31 2024-04-03 株式会社カクシン 管理サーバおよび人工知能チャットを利用した企画案生成システム

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101071418A (zh) * 2007-03-29 2007-11-14 腾讯科技（深圳）有限公司 聊天方法与系统
US20100082640A1 (en) * 2008-09-30 2010-04-01 Yahoo!, Inc. Guiding user moderation by confidence levels
CN106575503A (zh) * 2014-06-18 2017-04-19 微软技术许可有限责任公司 用于对话理解系统的会话上下文建模
CN106874292A (zh) * 2015-12-11 2017-06-20 北京国双科技有限公司 话题处理方法及装置
CN108415932A (zh) * 2018-01-23 2018-08-17 苏州思必驰信息科技有限公司 人机对话方法及电子设备

Family Cites Families (19)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP3844367B2 (ja) * 1994-05-17 2006-11-08 沖電気工業株式会社 音声情報通信システム
US6415257B1 (en) * 1999-08-26 2002-07-02 Matsushita Electric Industrial Co., Ltd. System for identifying and adapting a TV-user profile by means of speech technology
US8880406B2 (en) * 2011-03-28 2014-11-04 Epic Systems Corporation Automatic determination of and response to a topic of a conversation
CN104254851A (zh) * 2012-03-17 2014-12-31 海智网聚网络技术（北京）有限公司 用于向用户推荐内容的方法和系统
US20150120717A1 (en) * 2013-10-25 2015-04-30 Marketwire L.P. Systems and methods for determining influencers in a social data network and ranking data objects based on influencers
CN106326227A (zh) * 2015-06-17 2017-01-11 中兴通讯股份有限公司 一种提供在线客服的方法及装置
JP6601069B2 (ja) * 2015-09-01 2019-11-06 カシオ計算機株式会社 対話制御装置、対話制御方法及びプログラム
CN105183848A (zh) * 2015-09-07 2015-12-23 百度在线网络技术（北京）有限公司 基于人工智能的人机聊天方法和装置
CN105068661B (zh) * 2015-09-07 2018-09-07 百度在线网络技术（北京）有限公司 基于人工智能的人机交互方法和系统
CN105138671A (zh) * 2015-09-07 2015-12-09 百度在线网络技术（北京）有限公司 基于人工智能的人机交互的交互引导方法和装置
US20170091629A1 (en) * 2015-09-30 2017-03-30 Linkedin Corporation Intent platform
US10311862B2 (en) * 2015-12-23 2019-06-04 Rovi Guides, Inc. Systems and methods for conversations with devices about media using interruptions and changes of subjects
US10038787B2 (en) * 2016-05-06 2018-07-31 Genesys Telecommunications Laboratories, Inc. System and method for managing and transitioning automated chat conversations
JP6731326B2 (ja) * 2016-10-31 2020-07-29 ファーハット ロボティクス エービー 音声対話装置及び音声対話方法
US10489456B2 (en) * 2016-11-08 2019-11-26 International Business Machines Corporation Topic bridging determination using topical graphs
CN106653019B (zh) * 2016-12-07 2019-11-15 华南理工大学 一种基于用户注册信息的人机对话控制方法及系统
CN107066567B (zh) * 2017-04-05 2021-08-31 竹间智能科技（上海）有限公司 文字对话中基于话题侦测的用户画像建模方法及系统
US10044862B1 (en) * 2017-04-28 2018-08-07 International Business Machines Corporation Dynamic topic guidance in the context of multi-round conversation
US10552426B2 (en) * 2017-05-23 2020-02-04 International Business Machines Corporation Adaptive conversational disambiguation system

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101071418A (zh) * 2007-03-29 2007-11-14 腾讯科技（深圳）有限公司 聊天方法与系统
US20100082640A1 (en) * 2008-09-30 2010-04-01 Yahoo!, Inc. Guiding user moderation by confidence levels
CN106575503A (zh) * 2014-06-18 2017-04-19 微软技术许可有限责任公司 用于对话理解系统的会话上下文建模
CN106874292A (zh) * 2015-12-11 2017-06-20 北京国双科技有限公司 话题处理方法及装置
CN108415932A (zh) * 2018-01-23 2018-08-17 苏州思必驰信息科技有限公司 人机对话方法及电子设备

Also Published As

Publication number Publication date
JP2021512386A (ja) 2021-05-13
EP3745283A1 (en) 2020-12-02
US11735163B2 (en) 2023-08-22
US20200395000A1 (en) 2020-12-17
EP3745283A4 (en) 2021-09-22
CN108415932B (zh) 2023-12-22
JP7288446B2 (ja) 2023-06-07
CN108415932A (zh) 2018-08-17

Similar Documents

Publication Publication Date Title
WO2019144858A1 (zh) 人机对话方法及电子设备
CN108763495B (zh) 人机对话方法、系统、电子设备及存储介质
CN112071302B (zh) 计算代理的合成语音选择
CN108763494B (zh) 对话系统间的知识分享方法、对话方法及装置
Nguyen Potential effects of chatbot technology on customer support: A case study
Siddike et al. People’s interactions with cognitive assistants for enhanced performances
JP7488871B2 (ja) 対話推薦方法、装置、電子機器、記憶媒体ならびにコンピュータプログラム
US11347940B2 (en) Asynchronous role-playing system for dialog data collection
CN108846030B (zh) 访问官方网站的方法、系统、电子设备及存储介质
CN115062627B (zh) 用于基于人工智能的计算机辅助说服系统的方法和装置
US11978113B1 (en) Cross-device presentation with conversational user interface
Kokkodis et al. Learning to successfully hire in online labor markets
US12165633B2 (en) Intelligent training and education bot
Copulsky Do conversational platforms represent the next big digital marketing opportunity?
US11295731B1 (en) Artificial intelligence (AI) enabled prescriptive persuasion processes based on speech emotion recognition and sentiment analysis
AU2022339934A1 (en) Automated cognitive load-based task throttling
Weisswange et al. What could a social mediator robot do? Lessons from real-world mediation scenarios
Mohanty et al. Artificial intelligence transforming customer service management: Embracing the future
CN114757206A (zh) 咨询服务方法、系统及存储介质
Rydén et al. The leader’s strategic mindset: a key factor for AI success
US20240428260A1 (en) Artificial intelligence-powered contextual customer service through logic trees enhanced with mission and corporate values based interactions using generative ai prompts to control responses with brand characteristics and tone
Rayala Quantitative Analysis of Factors Affecting Trust Towards Voice Assistant Adoption
Esmailzadeh et al. Voice assistants and how they affect consumer behavior
Morrison Community Engagement and Climate Change: The Value of Social Networks and Community-Based Organizations
Smith Creating maximum value for the client: developing a new strategic role for market researchers

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 19743677

Country of ref document: EP

Kind code of ref document: A1

ENP Entry into the national phase

Ref document number: 2020537003

Country of ref document: JP

Kind code of ref document: A

NENP Non-entry into the national phase

Ref country code: DE

ENP Entry into the national phase

Ref document number: 2019743677

Country of ref document: EP

Effective date: 20200824