TWI806000B - Dialogue method and system based on complex task analysis - Google Patents

Abstract

A dialogue method based on complex task analysis comprises analyzing the dialogue with a user, extracting a keyword from the dialogue, determining which target complex task is related to the keyword and obtaining target subtasks having connections with the target complex task according to a knowledge graph, searching for matching product and service data for each target subtask, outputting an message corresponding to the product and service data matching each target subtask, receiving the next round of dialogue, and selectively adjusting one or more of the target complex tasks and target subtasks in the knowledge graph according to the next round of dialogue. The knowledge graph includes pre-stored complex tasks and pre-stored subtasks, wherein each pre-stored subtask has a connection with at least one of the pre-stored complex tasks, and the target complex task is one of the pre-stored complex tasks.

Description

Dialogue method and system based on complex task analysis

The present invention relates to a dialogue method, in particular to a dialogue method based on complex task analysis.

With the advancement of science and technology and the prosperity of the economy, people receive massive amounts of information every day, among which various news accounts for a relatively high proportion. However, the pace of modern life is very tight, and it is difficult for modern people to efficiently obtain useful and beneficial information in their busy daily life.

In recent years, chatbots have provided modern people with access to the information they seek. Like a virtual assistant, users can ask the chatbot questions such as weather and stock prices through voice or text, and the chatbot will search for the current weather, stock prices, etc. based on the query and respond to the user. However, this question-and-answer dialogue method is still insufficient in terms of information provision efficiency and comprehensiveness.

In view of the above, the present invention provides a dialogue method and system based on complex task analysis.

According to an embodiment of the present invention, the dialog method based on complex task analysis includes obtaining key words, judging that the key words are associated with the target complex task according to the knowledge map, and obtaining multiple target subtasks that have a connection relationship with the target complex task, for each The target subtask finds the matching product and service data, outputs the output message corresponding to the product service data matching each target subtask, receives the response message in response to the output message, and selectively adjusts the target complexity in the knowledge map according to the response message One or more of tasks and target subtasks. Wherein, the knowledge graph includes multiple pre-stored complex tasks and multiple pre-stored sub-tasks, each pre-stored sub-task has a connection relationship with at least one pre-stored complex task, and the target complex task is one of the pre-stored complex tasks.

According to an embodiment of the present invention, a dialog system based on complex task analysis includes a key word extraction device, a memory, a dialog device, and a natural language processing device, wherein the natural language processing device is connected to the key word retrieval device, the memory, and the dialog device. The key word retrieval device is used for obtaining key words. The memory stores the knowledge graph, wherein the knowledge graph includes multiple pre-stored complex tasks and multiple pre-stored sub-tasks, each pre-stored sub-task has a link relationship with at least one pre-stored complex task, and the target complex task is one of the pre-stored complex tasks one. The dialog device is used for providing an output message and receiving a response message in response to the output message. The natural language processing device is used to determine the key words associated with the target complex task based on the knowledge map and obtain multiple target subtasks that have a link relationship with the target complex task, and search for matching commodity service information for each target subtask, through the dialogue device Outputting output information corresponding to the commodity service data matching each target subtask, and selectively adjusting one or more of the target complex task and the target subtask in the knowledge graph according to the response message.

With the above structure, the dialogue method and system based on complex task analysis disclosed in this case can automatically find multiple associated subtasks based on single semantic information, and provide corresponding information, thereby improving the diversity of information provided, And enable users to efficiently obtain beneficial information.

The above description of the disclosure and the following description of the implementation are used to demonstrate and explain the spirit and principle of the present invention, and provide a further explanation of the patent application scope of the present invention.

The detailed features and advantages of the present invention are described in detail below in the implementation mode, and its content is enough to make any person familiar with the related art understand the technical content of the present invention and implement it accordingly, and according to the content disclosed in this specification, the scope of the patent application and the drawings , anyone skilled in the art can easily understand the purpose and advantages of the present invention. The following examples are to further describe the concept of the present invention in detail, but not to limit the scope of the present invention in any way.

Please refer to FIG. 1 . FIG. 1 is a functional block diagram of a dialog system based on complex task analysis according to an embodiment of the present invention, wherein the so-called complex task refers to a task associated with multiple subtasks. For example, "preparing for marriage" can be associated with subtasks such as "rental venue", "pre-purchasing dresses" and "wedding invitation design", so "preparing for marriage" is a kind of complex task. As shown in FIG. 1 , the dialog system 1 includes a key word extraction device 11, a dialog device 13, a memory 15, and a natural language processing device 17, wherein the natural language processing device 17 can be connected to the key word extraction device in a wired or wireless manner. Device 11 , dialog device 13 and memory 15 .

The keyword retrieval device 11 is used to obtain a keyword and provide it to the natural language processing device 17 . The natural language processing device 17 performs complex task analysis on the key words obtained by the key word extraction device 11 based on the data stored in the memory 15: judges the target complex task associated with the key word, and obtains a link with the target complex task A plurality of target subtasks of relationship; For each target subtask, search for one or more commodity service data matching; output message. After outputting the output message, the dialogue device 13 can receive a response message in response to the output message from the outside, and the natural language processing device 17 can selectively adjust the target complex tasks and target sub-tasks in the knowledge map according to the response message. One or more of the tasks.

The following is an illustration of one or more implementations of each device in the system. In one embodiment, the keyword extraction device 11 includes a speech input element and a speech recognizer connected to each other. The voice input component is, for example, a microphone and other receivers for receiving voice signals, wherein the so-called voice signal is formed, for example, by a sound waveform emitted by a user. In this embodiment, the input of a voice signal can be used as a triggering action. When the speech input element receives the input of the speech signal, the speech recognizer is triggered to perform speech recognition on the speech signal, so as to extract key words from the speech signal. In another embodiment, the keyword extraction device 11 includes input elements such as a keyboard, a mouse, or/and a touch panel, and a processor, wherein the input element can be triggered by actions such as tapping, clicking, and sliding to determine whether the key words have been received. The processor then analyzes the message to obtain the message corresponding to the action. In yet another implementation, when the webpage presenting an article (such as an Internet news article, a blog post, etc.) has been browsed for longer than a preset time, the keyword retrieval device 11 will obtain the keyword from the article , the key word is, for example, the title of the article, the hashtag of the article, and the like. Specifically, the keyword retrieval device 11 can be implemented by a processor in a computer that manages and controls webpages. When the webpage presenting an article is at the top of the display screen, the keyword retrieval device 11 judges that the article is being browsed. state and timing, and when the webpage presenting the article is at the top of the display screen for more than a preset time, key word extraction is executed.

As mentioned above, the dialogue device 13 can output output messages corresponding to the plurality of commodity service data, and can receive a response message in response to the output messages. Furthermore, the dialog device 13 includes an output element for outputting messages and an input element for inputting messages. The output element is, for example, a display, which can present information on the screen of the display in the form of text or pictures, or is, for example, a speaker, which can output information in the form of sound. The input elements are, for example, keyboards, mice, and touchpads, which can be triggered by actions such as tapping, clicking, and sliding to determine the received information corresponding to the actions. For another example, the input element may be a microphone and other receivers for receiving voice signals as response messages. To give another example, the input element may include a camera or both a radio and a camera, which can capture mouth images to identify mouth shapes, and then determine or assist voice recognition.

In another embodiment, the output element and the input element can be realized by a touch screen. In yet another embodiment, the output element and the input element can be implemented with a wired or wireless connection interface, and the wired or wireless connection interface can be connected to an external device (such as a mobile phone, a tablet, a personal computer, etc.) to transmit information corresponding to the multiple Text messages, e-mails, chat room messages, etc. for product and service information. Specifically, in addition to the output elements and input elements, the dialog device 13 may further include a natural language generation module, which may be realized by a central processing unit, a microcontroller, a programmable logic controller or other processors, or the above-mentioned processing software running on one of the devices. The natural language generation module can generate text or sound information in accordance with the natural language according to the commodity service data, and then output by the output component.

The memory 15 can be composed of one or more non-volatile storage media (such as flash memory, read-only memory, magnetic memory, etc.), and stores data required for complex task analysis. As shown in FIG. 1 , the memory 15 may include a knowledge graph 151 and a commodity service database 153 . The knowledge graph 151 may include multiple pre-stored complex tasks and multiple pre-stored sub-tasks, and each pre-stored sub-task has a connection relationship with at least one of the multiple pre-stored complex tasks. The knowledge graph 151 can be constructed based on multiple news corpora and blog corpora. Furthermore, the knowledge graph 151 can establish links between multiple complex tasks and multiple sub-tasks by analyzing words in news articles, blog posts, and the like. Wherein, the establishment method and examples of the knowledge graph 151 will be described later.

In addition to the knowledge graph 151 , the memory 15 may also include a commodity service database 153 . The commodity service database 153 stores a plurality of commodity service data, wherein each commodity service data may have a tag of a pre-stored subtask matched. Furthermore, in addition to the tags of the matched pre-stored subtasks, the commodity service data may also have tags of consumption demand patterns (eg, privileged patterns, reliable patterns, and expected patterns).

The natural language processing device 17 may include one or more central processing units, microcontrollers, programmable logic controllers or other processors. The natural language processing device 17 can perform complex task analysis and matching of commodity and service data on the key words obtained by the key word extraction device 11 . Furthermore, the analysis of complex tasks and the matching of commodity service data can be performed by different central processing units, microcontrollers, programmable logic controllers or other processors, or can be performed by multiple software.

In particular, the dialog device 13 can share an input element with the aforementioned keyword retrieval device 11 . The key word extraction device 11 can receive the user dialogue from the input element, extract key words therefrom, and then provide the key words to the natural language processing module 17 . The natural language processing module 17 performs complex task analysis on the key words based on the data stored in the memory 15 to determine which complex tasks and matching commodity service data the key words are associated with, and outputs the commodity service through the output component of the dialogue module 13 The message corresponding to the data. After outputting the message, the input component can receive the response message of the next round of dialogue from the outside, and the natural language processing device 17 can selectively adjust one of the target complex tasks and target subtasks in the knowledge map according to the response message. or more.

Furthermore, the knowledge graph 151 stored in the memory 15 can be constructed by two sub-modules in the natural language processing device 17 or other processors. The two submodules include a related subtask identification submodule (related subtask identification) and a complex task name generation submodule (complex task name generation), which can be implemented by two processors respectively, or by the same processing software running on the device. The relevant subtask detection submodule can use syntax analysis tools to cut sentences in news, microblogs (microblog), community question answering (community question answering, CQA) and other content into several words and phrases, and then use simple Candidate pairings for boilerplate rule extraction subtasks. The relevant subtask detection sub-module continues to collect a large amount of blog corpus to train the vocabulary vector (Word2Vector) with semantic concepts, and then expand it into a subtask vector (Subtask2Vector). The complex task name generation sub-module can use CQA and microblog containing related subtasks to generate complex task names, especially using the Topic-event-based complex task model (Topic-event-based complex task model) to generate subtasks complex task structure, and form a knowledge graph 151 with multiple complex task structures.

Please refer to FIG. 1, FIG. 2A and 2B together, wherein FIG. 2A is a schematic diagram of a knowledge map 151 of a dialog system 1 based on complex task analysis according to an embodiment of the present invention, and FIG. 2B is a schematic diagram according to another embodiment of the present invention. A schematic diagram of the knowledge map 151 of the dialogue system 1 based on complex task analysis shown in the embodiment. FIG. 2A exemplarily shows one of the pre-stored complex tasks CT1 in the knowledge graph 151 and the pre-stored sub-tasks ST1 , ST3 , and ST5 having a connection relationship with the pre-stored complex task CT1 . In the embodiment of FIG. 2A , the pre-stored complex task CT1 and its pre-stored sub-tasks ST1 , ST3 , and ST5 form a tree structure, wherein each pre-stored sub-task has a connection relationship with a single pre-stored complex task. The knowledge graph 151 in FIG. 2B presents a relatively complex distribution structure, in which a pre-stored sub-task can be linked to multiple pre-stored complex tasks at the same time. As shown in Figure 2B, the knowledge map 151 stores pre-stored complex tasks CT2 and CT4, wherein the pre-stored complex task CT2 has a link relationship with the pre-stored sub-tasks ST2, ST4, ST6 and ST8, and the pre-stored complex task CT4 is connected with the pre-stored sub-tasks ST2 and ST8. ST4 has a link relationship. In the embodiment shown in FIG. 2B , the pre-stored complex tasks CT2 and CT4 having a link relationship with the same pre-stored sub-task ST2 may respectively have historical usage ratios corresponding to the pre-stored sub-task ST2, wherein the historical usage ratio of the pre-stored complex task CT2 indicates Before the secondary complex task analysis, when the pre-stored sub-task ST2 is used as the input keyword, the probability of the pre-stored complex task CT2 being determined as the target complex task, and the historical usage ratio of the pre-stored complex task CT4 is the same.

In one embodiment, each pre-stored complex task in the knowledge graph 151 has a consumption demand pattern ratio; and in another embodiment, each pre-stored sub-task in the knowledge graph 151 has a consumption demand pattern ratio. Wherein, the consumption demand mode ratio is stored in the form of labels, for example. Further, the proportion of consumption demand patterns indicates the proportion among a plurality of consumption demand patterns, and the plurality of consumption demand patterns include a special-privilege pattern, a reliable pattern and a desired pattern. Specifically, consumer demand patterns are defined based on Maslow's hierarchy of needs. Furthermore, the definition of the privilege model is the same as that of Physiological needs in Maslow's hierarchy of needs theory, the definition of the reliable model is the same as that of safety needs in the hierarchy of needs theory, and the expectation model incorporates Belongingness and love needs, Esteem needs and Self-actualization needs in the hierarchy of needs theory. For example, in the case of the pre-stored complex task "Getting Married", the proportion of consumption demand mode can be as follows: the privilege mode accounts for 7.8%, the reliable mode accounts for 26.7%, and the expectation mode accounts for 65.5%. To give another example, in terms of the pre-stored subtask "buying a mobile phone", the proportion of consumption demand mode can be as follows: the privilege mode accounts for 51.1%, the reliable mode accounts for 38.5%, and the expectation mode accounts for 10.4%.

Please refer to FIG. 3 . FIG. 3 is a flowchart of a dialogue method based on complex task analysis according to an embodiment of the present invention. As shown in Figure 3, the dialogue method based on complex task analysis includes step S11: obtaining key words; step S12: according to the knowledge graph, judging that the key words are associated with the target complex task, and obtaining multiple targets that are linked to the target complex task Subtask; step S13: for each target subtask, search for one or more matching commodity service data; step S14: output the output message corresponding to one or more commodity service data matching each target subtask; step S15: Receive a response message in response to the output message; and Step S16: Selectively adjust one or more of the target complex task and the target subtask in the knowledge graph according to the response message.

Please refer to Fig. 1 and Fig. 3 together, the dialog method shown in Fig. 3 can be applicable to the dialog system 1 shown in Fig. 1, and the dialog system 1 of Fig. 1 is described exemplarily below and executes the dialog method of Fig. 3, but the present invention It is not limited that the dialog method in FIG. 3 is only applicable to the dialog system 1 shown in FIG. 1 .

In step S11 , the key word retrieval device 11 of the dialogue system 1 can obtain key words. Further, the key word extraction device 11 may be triggered by a trigger action associated with the semantic information, and extract key words from the semantic information. In one implementation, the semantic information is voice input, and the triggering behavior is the input voice signal; in another implementation, the semantic information is an article, and the triggering behavior is presenting the The webpage of the above-mentioned article has been browsed for more than a preset time; in yet another implementation, the semantic information is a dialog input by the user, and the triggering behavior is an input action.

In step S12 , the natural language processing device 17 of the dialog system 1 can determine that key words are associated with a target complex task according to the knowledge graph 151 , and obtain multiple target subtasks that are linked to the target complex task. Further, the natural language processing device 17 can judge whether the key vocabulary matches one of the multiple pre-stored complex tasks and multiple pre-stored sub-tasks stored in the knowledge map 151, which is called the conformer, and determines the target complex task accordingly. , and then use the pre-stored subtask that has a connection relationship with the target complex task as the target subtask. Furthermore, the natural language processing device 17 can determine whether there are words in the key vocabulary that are the same as the pre-stored complex tasks or pre-stored sub-tasks. When the matcher is one of the pre-stored complex tasks, the natural language processing device 17 will use the matcher as the target complex task; and when the matcher is one of the pre-stored sub-tasks, the natural language processing device 17 will match this matcher The matcher has a pre-existing complex task with a link relationship as a target complex task. After determining the target complex task, the natural language processing device 17 will then use the pre-stored subtasks linked to the target complex task as the target subtask according to the knowledge graph 151 .

Taking the knowledge graph 151 shown in FIG. 2A as an example, when the key word contains the word "preparing to get married", the natural language processing device 17 will judge that the key word is associated with the pre-stored complex task CT1, take it as the target complex task, and Obtain pre-stored sub-tasks ST1, ST3 and ST5 that are linked to this target complex task, and use these three pre-stored sub-tasks ST1, ST3 and ST5 as target sub-tasks; and when the keyword contains the word "venue rental" , the natural language processing device 17 will determine that the key words are associated with the pre-stored subtask ST1, and use the pre-stored complex task CT1 that has a connection relationship with the pre-stored sub-task ST1 as the target complex task, and then use the pre-stored sub-tasks ST1, ST3 and ST5 are used as target subtasks.

In addition, when the matcher is a pre-stored subtask, and there are multiple pre-stored complex tasks in the knowledge graph that have a connection relationship with the matcher, the natural language processing device 17 can select the highest sub-task among these pre-stored complex tasks (candidate complex tasks). History uses one of the scales as the target complex task. Taking the knowledge graph 151 shown in FIG. 2B as an example, when the key word contains the word "air ticket", the natural language processing device 17 will determine that the key word is associated with the pre-stored subtask ST2, and has a connection relationship with the pre-stored sub-task ST2 The pre-stored complex tasks CT2 and CT4 are included, so the natural language processing device 17 can determine which of the pre-stored complex tasks CT2 and CT4 has a higher historical usage ratio, and take it as the target complex task. In another implementation, the natural language processing device 17 can also output the pre-stored options of complex tasks CT2 and CT4 through the dialogue device 13 (such as outputting the button icon of the two options), and will be triggered (such as the clicked button icon) as the target complex task. Alternatively, the natural language processing device 17 may randomly select a pre-stored complex task CT2 or CT4 as the target complex task.

In step S13 , the natural language processing device 17 searches for one or more matching product and service materials for each target subtask. Further, the natural language processing device 17 may search the commodity service database 153 in the memory 15 for commodity service data matching the target subtask. As mentioned above, each commodity service data in the commodity service database 153 may have a tag of a pre-stored subtask matched. The natural language processing device 17 can search for tags to obtain commodity service information matching the target subtask. In addition, when there is no match in the commodity service database 153 , the natural language processing device 17 can further search for matching commodity service information from a search result snippet or a blog article.

In step S14 , the natural language processing device 17 outputs an output message corresponding to one or more commodity service data matching each target subtask through the dialogue device 13 . Wherein, the output message may be in the form of text or sound. For example, the natural language processing device 17 can output the brand, style, price and other information corresponding to the commodity service data or the order website in the form of chat room messages, short messages or emails. In steps S15 and S16, the natural language processing device 17 may receive a response message in response to the output message through the dialog device 13, and selectively adjust one or more of the target complex task and the target subtask in the knowledge graph according to the response message . In particular, the user can input a response message of whether the commodity service data selected by the system is appropriate through the dialogue device 13 .

For example, the output message corresponding to the commodity service data may include not only the brand, style, price and other information of the commodity service data or the order URL, but also an option of whether the recommendation of the target subtask corresponding to the commodity service is needed. When the natural language processing device 17 receives an unnecessary response message, the natural language processing device 17 can save and delete the knowledge map of the link relationship between the target subtask and the target complex task as a part of providing the unnecessary response message The user's dedicated knowledge graph provides the function of customizing the knowledge graph. When the natural language processing device 17 receives a specific number of unnecessary response messages for a specific target subtask, it will delete the connection relationship between the target subtask and the target complex task in the knowledge graph 151 . In addition, when the natural language processing device 17 receives unnecessary messages corresponding to multiple target subtasks at the same time, the natural language processing device 17 may determine that the target complex task determined in the previous step S12 is wrong, and re-determine the target complex task , and reduce the historical usage ratio of the originally selected target complex task.

In addition to searching for matching product and service data with the tags of the pre-stored subtasks, the natural language processing device 17 can also search for product and service data matching the target subtask based on the consumption demand pattern ratio, so that the provided product and service data can be more Close to the psychology of consumers. As mentioned above, the pre-stored complex tasks or pre-stored sub-tasks in the knowledge graph 151 may have consumption demand pattern proportions. In the embodiment in which each pre-stored subtask has a consumption demand pattern ratio, after the natural language processing device 17 finds the commodity service data with the target subtask label, it can then base the target subtask with the highest proportion of consumption demand pattern ratios. Consumer demand patterns to filter commodity and service information. For example, when the proportion of reliable patterns among the consumer demand patterns of the target subtask is the highest, the natural language processing device 17 will select the commodity service data that has both the target subtask label and the reliable pattern label. Alternatively, the natural language processing device 17 may determine the order of presenting commodity service information according to the proportion of each consumption demand mode in the proportion of consumption demand modes.

In an embodiment where each pre-stored complex task has a consumption demand pattern ratio, the natural language processing device 17 can set each target subtask to have a consumption demand pattern ratio that the target complex task has, and output the output corresponding to each The output information of the consumption demand pattern ratio of each target subtask. The natural language processing device 17 can receive a response message in response to the output message including the consumption demand pattern ratio through the dialogue device 13, and selectively adjust the consumption demand pattern ratio of one or more of the target subtasks according to the response message. In particular, the natural language processing device 17 can target the consumption demand pattern ratio of the target complex task as the preset consumption demand pattern ratio of each target subtask, and then provide the user with the dialogue device 13 to adjust the consumption demand pattern ratio of each target subtask pipeline. Then, the natural language processing device 17 screens the commodity service data according to the proportion of the consumption demand pattern of the target subtask after selective adjustment, and the way of screening or presenting the commodity service data is as in the front row. Each pre-stored subtask has a consumption demand pattern proportion. As described in the embodiment, it will not be repeated here. Or, the natural language processing device 17 can directly search for the commodity service data with the target subtask label and the label with the highest proportion in the consumption demand mode proportion based on the consumption demand mode proportion of the target complex task, or according to the consumption demand mode The ratio is used to determine the presentation order of commodity service data. By providing appropriate product and service information based on the above psychological needs, the chances of successful advertisement recommendation and product service sales can be improved.

With the above structure, the dialogue method and system based on complex task analysis disclosed in this case can automatically find multiple associated subtasks based on single semantic information, and provide corresponding information, thereby improving the diversity of information provided, And enable users to efficiently obtain beneficial information.

Although the present invention is disclosed by the aforementioned embodiments, they are not intended to limit the present invention. Without departing from the spirit and scope of the present invention, all changes and modifications are within the scope of patent protection of the present invention. For the scope of protection defined by the present invention, please refer to the appended scope of patent application.

1: Dialogue system 11:Key word extraction device 13: dialogue device 15: Memory 151: Knowledge Graph 153:Commodity service database 17: Natural language processing device CT1, CT2, CT4: Pre-store complex tasks ST1～ST6, ST8: pre-stored subtasks

FIG. 1 is a functional block diagram of a dialogue system based on complex task analysis according to an embodiment of the present invention. FIG. 2A is a schematic diagram of a knowledge graph of a dialogue system based on complex task analysis according to an embodiment of the present invention. FIG. 2B is a schematic diagram of a knowledge graph of a dialogue system based on complex task analysis according to another embodiment of the present invention. FIG. 3 is a flowchart of a dialogue method based on complex task analysis according to an embodiment of the present invention.

1: Dialogue system

11:Key word extraction device

13: dialogue device

15: Memory

151: Knowledge Graph

153:Commodity service database

17: Natural language processing device

Claims (11)

A dialogue method based on complex task analysis, suitable for a dialogue system, comprising: obtaining a key word; judging that the key word is associated with a target complex task according to a knowledge map, and obtaining multiple links related to the target complex task A target subtask; for each of the target subtasks, search for one or more matching commodity service data; output an output message corresponding to the one or more commodity service data matching each of the target subtasks ; receiving a response message in response to the output message; and selectively adjusting the target complex task and one or more of the target subtasks in the knowledge graph according to the response message; wherein the knowledge graph includes multiple pre-stored A complex task and multiple pre-stored sub-tasks, each of the pre-stored sub-tasks has a link relationship with at least one of the pre-stored complex tasks, and the target complex task is one of the pre-stored complex tasks, Selectively adjusting one or more of the target complex task and the target subtasks in the knowledge graph according to the response message includes: when the response message is an unnecessary response message, saving and deleting the targets Another knowledge graph of the link relationship between one or more of the subtasks and the target complex task; and when the number of the response message is that the response message does not need to reach a specific number, delete the knowledge graph in the knowledge graph A linkage relationship between the target complex task and one or more of the target subtasks. The dialogue method as described in claim 1, wherein each of the pre-stored complex tasks in the knowledge graph has a consumption demand pattern ratio, the output message is a first output message, and the response message is a first response message, And the dialogue method further includes: setting each of the target subtasks to have the proportion of the consumption demand pattern of the target complex task, and outputting a second output message corresponding to the proportion of the consumption demand pattern of each of the target subtasks ; and receiving a second response message in response to the second output message, and selectively adjusting the proportion of the consumption demand pattern of one or more of the target subtasks according to the second response message; Wherein for each of the target subtasks, the step of finding the one or more matching commodity service data is based on the proportion of the consumption demand pattern of the target subtask. The dialogue method as described in claim 1, wherein each of the pre-stored subtasks in the knowledge graph has a consumption demand pattern ratio, and for each of the target subtasks, search for the matching one or more commodity services The steps of the profile are based on the proportion of the consumption demand pattern of the target subtask. The dialog method according to claim 2 or 3, wherein the consumption demand mode ratio indicates the proportion among a plurality of consumption demand modes, and the consumption demand modes include a privileged mode, a reliable mode and an expected mode. The dialogue method as described in claim 1, wherein according to the knowledge graph, judging that the key word is associated with the target complex task includes: judging that the key word matches the pre-stored complex tasks and the pre-stored sub-tasks in the knowledge graph When the matcher is one of the pre-stored complex tasks, the matcher is used as the target complex task; and when the matcher is one of the pre-stored sub-tasks, the Among the complex tasks, those who have a link relationship with the corresponding person select the target complex task. The dialogue method as described in claim 5, wherein those pre-stored complex tasks that have a link relationship with one of the pre-stored sub-tasks include a plurality of candidate complex tasks, and each of the candidate complex tasks has a historical use ratio, the step of selecting the target complex task from those who have a connection relationship with the matchers of the pre-stored subtasks from the pre-stored complex tasks includes: selecting one of the candidate complex tasks with the highest historical usage ratio as The target complex task. The dialogue method as described in claim 1, wherein obtaining the key words includes: receiving a voice signal, and obtaining the key words from the voice signal. The dialogue method according to claim 1, wherein obtaining the key word includes: when it is determined that an article is in a browsed state for more than a preset time, being triggered to obtain the key word from the article. A dialogue system based on complex task analysis, comprising: a key word extraction device for obtaining a key word; a memory for storing a knowledge map, wherein the knowledge map includes multiple pre-stored complex tasks and multiple pre-stored sub-tasks , each of these pre-stored subtasks is associated with the At least one of the pre-stored complex tasks has a link relationship, and the target complex task is one of the pre-stored complex tasks; a dialogue device is used to provide an output message and receive a response to the output message Response message; and a natural language processing device connected to the keyword retrieval device, the memory and the dialogue device, and used to judge that the keyword is associated with a target complex task based on the knowledge map, and obtain the target complex task The complex task has a plurality of target subtasks that are connected, and searches for one or more matching commodity data or service data for each of these target subtasks, and outputs the matching target subtasks through the dialogue device. The output message corresponding to one or more commodity service data, and selectively adjust the target complex task and one or more of the target sub-tasks in the knowledge graph according to the response message, wherein the natural language processing The device selectively adjusts one or more of the target complex task and the target subtasks in the knowledge graph according to the response message, including: when the response message is an unnecessary response message, save and delete the targets another knowledge graph of linkages between one or more of the subtasks and the target complex task; and When the number of the response messages being the unnecessary response messages reaches a specific number, delete the connection relationship between the target complex task and one or more of the target sub-tasks in the knowledge graph. The dialog system as described in claim 9, wherein each of the pre-stored complex tasks in the knowledge graph has a consumption demand pattern ratio, the output message is a first output message, and the response message is a first response message, Wherein the natural language processing device further sets each of the target subtasks to have the proportion of the consumption demand pattern of the target complex task, and outputs a second output message corresponding to the proportion of the consumption demand pattern of each of the target subtasks , and receive a second response message in response to the second output message, and selectively adjust the proportion of the consumption demand pattern of one or more of the target subtasks according to the second response message, and the natural language For each of the target subtasks performed by the processing device, searching for the one or more matching commodity service data is based on the proportion of the consumption demand pattern of the target subtask. The dialog system as described in claim 9, wherein each of the pre-stored subtasks in the knowledge graph has a consumption demand pattern ratio, and the natural language processing device executes for each of the target subtasks, finds a matching The one or more commodity service data are based on the proportion of the consumption demand pattern of the target subtask.

Images