A method and apparatus for recommending questions

Abstract

The present invention discloses a method and apparatus for recommending questions, relating to the field of computer technology. A specific embodiment of the method includes: performing intent recognition on the session information input by a user to obtain an intent recognition result; matching at least one recommended question according to the intent recognition result and a node relationship graph; wherein the attribute of each node includes an intent and a question; and pushing the at least one recommended question to the user. This embodiment can solve the technical problem of manually maintaining a recommended question list.

Description

Technical Field

[0001] The present invention relates to the field of computer technology, and in particular, to a method and apparatus for recommending questions. Background Art

[0002] The main response process of an intelligent dialogue system is intent recognition (NLU) and classification response; for example, when the user says "I want to modify the order", the intent recognition result is "modify the order", and the classification response matches the corresponding solution to respond "The order has been shipped and entered the distribution process, and the order attributes cannot be modified". The intelligent response system not only provides answers to the user's questions, but also recommends other potential questions that the user may want to ask. The user obtains the answers by clicking on the recommended questions, so as to close the loop of potential questions related to the user's intent in a single conversation.

[0003] The recommended questions are divided into two types: one is the intent-fuzzy recommended questions. At this time, no response can be made, and questions must be recommended to guide the user to provide accurate intent; the other is the recommended potential questions with answers. At this time, the intent is accurate, but the user may continue to ask questions about this answer. At this time, recommended questions are appended after the answer to guide the user to consult potential questions and improve the user experience.

[0004] In the process of implementing the present invention, the inventors found that there are at least the following problems in the prior art:

[0005] When the business scenario changes, more questions can only be recommended by modifying the question list, and the question content and order can only be manually maintained, resulting in high maintenance costs. Summary of the Invention

[0006] In view of this, embodiments of the present invention provide a method and apparatus for recommending questions to solve the technical problem of manually maintaining the recommended question list.

[0007] To achieve the above object, according to one aspect of the embodiments of the present invention, a method for recommending questions is provided, including:

[0008] Performing intent recognition on the session information input by the user to obtain an intent recognition result;

[0009] Matching at least one recommended question according to the intent recognition result and the node relationship graph; wherein, the attributes of each node include intent and question;

[0010] Pushing the at least one recommended question to the user.

[0011] Optionally, performing intent recognition on the session information input by the user to obtain an intent recognition result includes:

[0012] Performing intent recognition on the session information input by the user to obtain multiple fuzzy intents or one accurate intent.

[0013] Optionally, if the intention recognition result is multiple fuzzy intentions;

[0014] Then, according to the intention recognition result and the node relationship graph, at least one recommended question is matched, including:

[0015] Match the node questions corresponding to each fuzzy intention respectively according to the node relationship graph;

[0016] Calculate the similarity between each node question and the session information respectively, and screen out the target node question with the highest similarity;

[0017] According to the target node question and the node relationship graph, at least one recommended question is matched.

[0018] Optionally, according to the target node question and the node relationship graph, at least one recommended question is matched, including:

[0019] Take the node where the target node question is located as the starting node, and match at least one destination node according to the node relationship graph;

[0020] Sort the at least one destination node based on the relationship attributes between the starting node and the destination node, screen out at least one target destination node, and use the questions in the at least one target destination node as the recommended questions.

[0021] Optionally, if the intention recognition result is an accurate intention;

[0022] Then, according to the intention recognition result and the node relationship graph, at least one recommended question is matched, including:

[0023] Take the node where the accurate intention is located as the starting node, and match at least one destination node according to the node relationship graph;

[0024] Sort the at least one destination node based on the relationship attributes between the starting node and the destination node, screen out at least one target destination node, and use the questions in the at least one target destination node as the recommended questions.

[0025] Optionally, the relationship attributes between the starting node and the destination node include weight value and / or heat value.

[0026] Optionally, after recommending the at least one recommended question to the user, it further includes:

[0027] Update the node relationship graph according to the click result of the user on the at least one recommended question and the accurate intention.

[0028] Optionally, update the node relationship graph according to the user's click results on the at least one recommended question and the accurate intention, including:

[0029] Obtain the clicked recommended question from the user's click results on the at least one recommended question;

[0030] Take the node where the clicked recommended question is located as the destination node, and take the node where the accurate intention is located as the starting node, and update the relationship attribute between the starting node and the destination node.

[0031] Optionally, update the node relationship graph according to the user's click results on the at least one recommended question and the accurate intention, including:

[0032] If the user's click result is none and the user expresses the current intention, determine whether the current intention exists in the node relationship graph;

[0033] If not, create a new destination node, and the attributes of the destination node include the current intention and its corresponding question;

[0034] If so, take the node where the current intention is located as the destination node;

[0035] Take the node where the accurate intention is located as the starting node, and establish the relationship between the starting node and the destination node;

[0036] Update the relationship attribute between the starting node and the destination node.

[0037] Optionally, update the relationship attribute between the starting node and the destination node, including:

[0038] Obtain all destination nodes having an associated relationship with the starting node, and calculate the sum of the heat values between the starting node and all the destination nodes;

[0039] According to the heat value between the starting node and the destination node, and the sum of the heat values between the starting node and all the destination nodes, calculate the heat percentage between the starting node and the destination node;

[0040] Determine the heat coefficient between the starting node and the destination node according to the number of relationships pointed to by the destination node;

[0041] Calculate the weight between the starting node and the destination node according to the heat percentage and the heat coefficient between the starting node and the destination node.

[0042] In addition, according to another aspect of the embodiments of the present invention, a device for recommending questions is provided, including:

[0043] An identification module, configured to perform intent identification on the session information input by the user to obtain an intent identification result;

[0044] A matching module, configured to match at least one recommended question according to the intent identification result and the node relationship graph; wherein, the attribute of each node includes an intent and a question;

[0045] A recommendation module, configured to push the at least one recommended question to the user.

[0046] Optionally, the identification module is further configured to:

[0047] Perform intent identification on the session information input by the user to obtain multiple fuzzy intents or one accurate intent.

[0048] Optionally, if the intent identification result is multiple fuzzy intents;

[0049] Then the matching module is further configured to:

[0050] Match the node questions corresponding to each fuzzy intent respectively according to the node relationship graph;

[0051] Calculate the similarity between each node question and the session information respectively, and screen out the target node question with the highest similarity;

[0052] Match at least one recommended question according to the target node question and the node relationship graph.

[0053] Optionally, the matching module is further configured to:

[0054] Take the node where the target node question is located as the starting node, and match at least one destination node according to the node relationship graph;

[0055] Sort the at least one destination node based on the relationship attribute between the starting node and the destination node, screen out at least one target destination node, and use the questions in the at least one target destination node as recommended questions.

[0056] Optionally, if the intent identification result is one accurate intent;

[0057] Then the matching module is further configured to:

[0058] Take the node where the accurate intent is located as the starting node, and match at least one destination node according to the node relationship graph;

[0059] Sort the at least one destination node based on the relationship attributes between the starting node and the destination node, filter out at least one target destination node, and use the problems in the at least one target destination node as recommended problems.

[0060] Optionally, the relationship attributes between the starting node and the destination node include weight and / or heat value.

[0061] Optionally, it further includes an update module for:

[0062] After recommending the at least one recommended problem to the user, update the node relationship graph according to the click results of the user on the at least one recommended problem and the accurate intention.

[0063] Optionally, the update module is further used for:

[0064] Obtain the clicked recommended problem from the click results of the user on the at least one recommended problem;

[0065] Take the node where the clicked recommended problem is located as the destination node, and take the node where the accurate intention is located as the starting node, and update the relationship attributes between the starting node and the destination node.

[0066] Optionally, the update module is further used for:

[0067] If the click result of the user is none and the user expresses the current intention, judge whether the current intention exists in the node relationship graph;

[0068] If not, create a new destination node, and the attributes of the destination node include the current intention and its corresponding problem;

[0069] If so, take the node where the current intention is located as the destination node;

[0070] Take the node where the accurate intention is located as the starting node, and establish the relationship between the starting node and the destination node;

[0071] Update the relationship attributes between the starting node and the destination node.

[0072] Optionally, the update module is further used for:

[0073] Obtain all destination nodes having an associated relationship with the starting node, and calculate the sum of the heat values between the starting node and all the destination nodes;

[0074] Calculate the heat percentage between the departure node and the destination node based on the heat value between the departure node and the destination node, and the sum of the heat values between the departure node and all destination nodes;

[0075] Determine the heat coefficient between the departure node and the destination node according to the number of relationships in which the destination node is pointed to;

[0076] Calculate the weight between the departure node and the destination node according to the heat percentage and the heat coefficient between the departure node and the destination node.

[0077] According to another aspect of the embodiments of the present invention, there is also provided an electronic device, including:

[0078] One or more processors;

[0079] A storage device for storing one or more programs,

[0080] When the one or more programs are executed by the one or more processors, the one or more processors implement the method described in any of the above embodiments.

[0081] According to another aspect of the embodiments of the present invention, there is also provided a computer-readable medium, on which a computer program is stored, and when the program is executed by a processor, the method described in any of the above embodiments is implemented.

[0082] One embodiment of the above invention has the following advantages or beneficial effects: Because the technical means of performing intent recognition on the session information input by the user to obtain an intent recognition result, and matching at least one recommended question according to the intent recognition result and the node relationship graph is adopted, the technical problem of manually maintaining the recommended question list in the prior art is overcome. The embodiments of the present invention match at least one recommended question through the node relationship graph, thereby pushing the recommended question to the user, and automatically dynamically updating the node relationship graph according to the user's click result and input content, which can not only improve the accuracy of subsequent matching, but also does not require the operation and maintenance personnel to manually maintain the recommended question list. The embodiments of the present invention can improve the operation and maintenance efficiency, save 3% of the operation efficiency of personnel, and also improve the user experience and increase the problem-solving rate by 5%.

[0083] The further effects of the above non-conventional optional methods will be described in conjunction with the specific embodiments below. Description of the Drawings

[0084] The drawings are used to better understand the present invention and do not constitute an improper limitation of the present invention. Among them:

[0085] Figure 1Schematic diagram of the main process of the method for recommending questions according to an embodiment of the present invention;

[0086] Figure 2 Schematic diagram of the node relationship diagram according to an embodiment of the present invention;

[0087] Figure 3 Schematic diagram of the main process of the method for recommending questions according to a reference embodiment of the present invention;

[0088] Figure 4 Schematic diagram of the main process of the method for recommending questions according to another reference embodiment of the present invention;

[0089] Figure 5 Schematic diagram of the main process of the method for recommending questions according to still another reference embodiment of the present invention;

[0090] Figure 6 Schematic diagram of the main process of the method for recommending questions according to still another reference embodiment of the present invention;

[0091] Figure 7 Schematic diagram of the main modules of the device for recommending questions according to an embodiment of the present invention;

[0092] Figure 8 Exemplary system architecture diagram to which the embodiments of the present invention can be applied;

[0093] Figure 9 Schematic diagram of the structure of a computer system of a terminal device or a server suitable for implementing the embodiments of the present invention. Detailed implementation manners

[0094] The following makes an explanation of exemplary embodiments of the present invention with reference to the accompanying drawings. Various details of the embodiments of the present invention are included to facilitate understanding, and they should be considered merely exemplary. Therefore, those of ordinary skill in the art should recognize that various changes and modifications can be made to the embodiments described herein without departing from the scope and spirit of the present invention. Similarly, for clarity and conciseness, descriptions of well-known functions and structures are omitted in the following description.

[0095] The recommended questions are divided into two types: one is the recommended question with ambiguous intent. At this time, it cannot be answered, and the recommended question must be used to guide the user to provide accurate intent; the other is the recommended potential question with an answer. At this time, the intent is precise, but the user may continue to ask questions about this answer. At this time, the recommended question is appended after the answer to guide the user to consult potential questions and improve the user experience.

[0096] For example:

[0097]

[0098] The main implementation methods for the problem of fuzzy intention recommendation are as follows: 1. Through NLU (Natural Language Understanding) intention recognition, if the intention classification value in the recognition result is lower than the intention accuracy threshold, then output the fuzzy intention and the list of fuzzy intentions; 2. The response system obtains the fuzzy classification list and assembles the fuzzy classification standard questions to form recommended questions.

[0099] However, with the development of the business, when the business scenario changes and more refined intention questions need to be recommended when the intention is fuzzy, it can only be achieved by retraining the model. However, the training model has a long cycle and it is difficult to guarantee the effect. The business changes quickly and unstably. At this time, this mode cannot meet the business requirements. For example, when modifying the order classification to be fuzzy, recommended questions such as "I want to modify the order address" and "I want to modify the order phone number" are required. Such refined classification questions introduced due to business refinement often change, and it is inefficient to add new classifications through model training.

[0100] The implementation method for recommending potential questions with answers is as follows: Currently, it completely depends on the operation personnel to sort out the business, sort out the potential recommended questions related to the classification, and then configure the potential questions into the classification answers. When the user's question hits this classification, the response system responds with both the business answer and the recommended question.

[0101] However, the current implementation method can only manually configure potential questions, and the answer list is fixed after configuration and cannot be dynamically adjusted according to user clicks. For example, in the modified order classification answer, the answer + recommended question is configured. Whether it is the question content or the order, only the recommended question list can be manually maintained, and the maintenance cost is high; moreover, it is necessary to extract and adjust the content and order based on the click popularity monitored online, and the maintenance efficiency is low.

[0102] To solve the above problems existing in the prior art, an embodiment of the present invention provides a method for recommending questions to achieve dynamic configuration of recommended questions.

[0103] Figure 1 It is a schematic diagram of the main process of the method for recommending questions according to an embodiment of the present invention. As an embodiment of the present invention, as Figure 1 shown, the method for recommending questions may include:

[0104] Step 101, perform intention recognition on the session information input by the user to obtain an intention recognition result.

[0105] Before answering the user's question, first perform intent recognition on the session information input by the user to obtain the intent recognition result, and then recommend questions to the user. Optionally, natural language understanding can be used to perform intent recognition on the session information input by the user to obtain multiple fuzzy intents or one accurate intent. It should be noted that if the user's intent is fuzzy, the intent recognition result is multiple fuzzy intents; if the user's intent is accurate, the intent recognition result is one accurate intent. In addition, in the intent recognition result, in addition to the intent, it can also include the questions corresponding to each intent (i.e., standard question forms).

[0106] Step 102, match at least one recommended question according to the intent recognition result and the node relationship graph.

[0107] According to the intent recognition result obtained in step 101, match at least one recommended question from the node relationship graph in the graph database. Among them, the attributes of each node include intent and question. The node relationship graph stores the relationships between nodes, and the node relationship graph can be updated in real time. Figure 2 Schematic diagram of the node relationship graph according to an embodiment of the present invention, as Figure 2 shown, the node relationships are all one-way relationships, and each relationship has two nodes, namely the starting node and the destination node; each node includes two attributes, namely intent and question (i.e., the standard question form corresponding to the intent). Optionally, the relationship attribute between the starting node and the destination node can include a weight value, or can include a heat value, or can include both a weight value and a heat value. Optionally, when initializing the node relationship graph, the intent switching data of each existing user, the click data of recommended questions, etc. can be extracted from the online session, so as to calculate the weight value and heat value between nodes and obtain the node relationship graph.

[0108] Optionally, if the intent recognition result is multiple fuzzy intents, matching at least one recommended question according to the intent recognition result and the node relationship graph includes: respectively matching the node questions corresponding to each fuzzy intent according to the node relationship graph; respectively calculating the similarity between each node question and the session information, and screening out the target node question with the highest similarity; according to the target node question and the node relationship graph, matching at least one recommended question. When the user expresses a fuzzy intent, obtain multiple fuzzy intents from the fuzzy list, and match the node questions corresponding to each fuzzy intent from the node relationship graph, as Figure 2As shown, each fuzzy intent and its corresponding node question are on the same node. Then, a text similarity algorithm is used to calculate the similarity between each node question and the question input by the user, such as DSSM (Deep Semantic Similarity Model), so as to screen out the target node question with the highest similarity. After determining the target node question, at least one recommended question is matched according to the target node question and the node relationship graph. It should be noted that in the case of fuzzy intent, the question input by the user is not determined, so there is no answer, and only potential questions related to the expression are recommended to the user.

[0109] Optionally, matching at least one recommended question according to the target node question and the node relationship graph includes: taking the node where the target node question is located as the starting node, and matching at least one destination node according to the node relationship graph; sorting the at least one destination node based on the relationship attribute between the starting node and the destination node, screening out at least one target destination node, and taking the questions in the at least one target destination node as recommended questions. After determining the target node question, take the node where the target node question is located as the starting node, and find all the relationships starting from this node in the node relationship graph, so as to match each destination node. Then, based on the weight value and / or heat value between the starting node and each destination node, sort the at least one destination node, screen out the top N target destination nodes, and take the questions in the N target destination nodes as recommended questions. Optionally, it can be sorted in descending order according to the size of the weight value. If the weight values are the same, it is sorted in descending order according to the heat value. It can also be sorted in descending order according to the size of the heat value. If the heat values are the same, it is sorted in descending order according to the weight value. It is also possible to set the weights of the weight value and the heat value respectively, and perform a descending order arrangement by means of weighted summation.

[0110] For example, if the question input by the user is "I wrote the address wrong", the intent recognition result is fuzzy intent, including three fuzzy intents: modifying the order, modifying the order address, and whether the order can be modified. Figure 2As shown, three node problems are matched from the node relationship graph according to these three fuzzy intents: I want to modify the order, I want to modify the delivery address of the order, and I want to inquire whether the order can be modified. The text similarity algorithm is used to calculate the similarity between "I wrote the wrong address" and these three node problems, and the node problem with the highest similarity is selected, that is, "I want to modify the delivery address of the order". Then, the node where this node problem is located ("Modify order address / I want to modify the delivery address of the order") is used as the starting node, and four destination nodes are matched from the node relationship graph. Based on the weight and heat value, the top three problems are selected from these four destination nodes as recommended problems, that is, I want to modify the phone number of the order, I want to cancel the order, and what is the contact information of the delivery person.

[0111]

[0112] Optionally, if the intent recognition result is an accurate intent, at least one recommended problem is matched according to the intent recognition result and the node relationship graph, including: using the node where the accurate intent is located as the starting node, and matching at least one destination node according to the node relationship graph; sorting the at least one destination node based on the relationship attribute between the starting node and the destination node, screening out at least one target destination node, and using the problems in the at least one target destination node as recommended problems. When the user's expressed intent is precise, the node where the accurate intent is located is used as the starting node, and all relationships with this node as the starting node are found from the node relationship graph, so as to match each destination node. Then, based on the weight and / or heat value between the starting node and each destination node, the at least one destination node is sorted, the top N target destination nodes are screened out, and the problems in the N target destination nodes are used as recommended problems. Optionally, it can be sorted in descending order according to the size of the weight. If the weights are the same, it is sorted in descending order according to the heat value. It can also be sorted in descending order according to the size of the heat value. If the heat values are the same, it is sorted in descending order according to the weight. It is also possible to set the weights of the weight and the heat value respectively, and perform a descending order arrangement by means of weighted summation. When the user's expressed intent is precise, there is no need to filter the node problems through similarity, but directly push the answer + recommended problems.

[0113] For example, if the user's input question is "I want to modify the order address", the intent recognition result is an accurate intent, that is, modify the order address. Using the node where this accurate intent is located ("Modify order address / I want to modify the delivery address of the order") as the starting node, four destination nodes are found from the node relationship graph. Then, based on the weight and heat value, the top three problems are selected from these four destination nodes as recommended problems, that is, I want to modify the phone number of the order, I want to cancel the order, and what is the contact information of the delivery person.

[0114]

[0115]

[0116] It should be noted that, whether the intention is ambiguous or accurate, if the intention does not exist in the node relationship graph, recommended questions are matched through the recommended question list in the existing method.

[0117] Step 103: Push the at least one recommended question to the user.

[0118] After matching at least one recommended question, the server sends the at least one recommended question to the user terminal. The user may click on one of the recommended questions. When the user terminal reports the click result of the user to the server. Optionally, after recommending the at least one recommended question to the user, it further includes: updating the node relationship graph according to the click result of the user on the at least one recommended question and the accurate intention. If the intention recognition result is an accurate intention, the node relationship graph can be updated according to the click result of the user on the at least one recommended question and the accurate intention. If the intention recognition result is multiple ambiguous intentions, the node relationship graph cannot be updated. It should be noted that whether the user clicks or inputs, the node relationship graph can be updated according to the click result and input content of the user.

[0119] If the user clicks on a certain recommended question, updating the node relationship graph according to the click result of the user on the at least one recommended question and the accurate intention includes: obtaining the clicked recommended question from the click result of the user on the at least one recommended question; taking the node where the clicked recommended question is located as the destination node, and taking the node where the accurate intention is located as the starting node, and updating the relationship attribute between the starting node and the destination node. The embodiment of the present invention automatically performs dynamic update of the node relationship graph according to the accurate intention and the click result of the user, which can not only improve the accuracy of subsequent matching, but also does not require the operation and maintenance personnel to manually maintain the recommended question list.

[0120] If the user does not click on any of the recommended questions but directly enters an accurate current intention, then based on the click result of the user on the at least one recommended question and the accurate intention, update the node relationship graph, including: if the click result of the user is none and the user expresses the current intention, then determine whether the current intention exists in the node relationship graph; if not, then create a new destination node, and the attributes of the destination node include the current intention and its corresponding question; if so, then use the node where the current intention is located as the destination node; use the node where the accurate intention is located as the starting node, establish the relationship between the starting node and the destination node; update the relationship attributes between the starting node and the destination node. In this case, first determine whether the current intention entered by the user exists in the node relationship graph. If it exists, then use the node where the current intention is located as the destination node. If it does not exist, then create a new destination node; and use the node where the accurate intention is located as the starting node, and update the relationship attributes between the starting node and the destination node.

[0121] According to the various embodiments described above, it can be seen that the present invention solves the technical problem of manually maintaining a recommended question list in the prior art by means of intention recognition of the session information input by the user to obtain an intention recognition result, and matching at least one recommended question according to the intention recognition result and the node relationship graph. The embodiments of the present invention match at least one recommended question through the node relationship graph, thereby pushing the recommended question to the user, and automatically dynamically updating the node relationship graph according to the click result and input content of the user, which can not only improve the accuracy of subsequent matching, but also does not require the operation and maintenance personnel to manually maintain the recommended question list. The embodiments of the present invention can improve the operation and maintenance efficiency, save 3% of the operation efficiency of personnel, and also improve the user experience and increase the problem-solving rate by 5%.

[0122] Figure 3 It is a schematic diagram of the main process of a method for recommended questions according to a reference embodiment of the present invention. As another embodiment of the present invention, the method for recommended questions may include the following steps:

[0123] Step 301, perform intention recognition on the session information input by the user to obtain an accurate intention.

[0124] Step 302, determine whether the accurate intention exists in the node relationship graph; if so, then execute step 303; if not, then end.

[0125] Step 303, use the node where the accurate intention is located as the starting node, and match at least one destination node according to the node relationship graph.

[0126] Step 304: Sort the at least one destination node based on the relationship attributes between the departure node and the destination node, screen out at least one target destination node, and use the questions in the at least one target destination node as recommended questions.

[0127] Step 305: Push the at least one recommended question to the user.

[0128] Step 306: Update the node relationship graph according to the click results of the user on the at least one recommended question and the accurate intention.

[0129] In addition, the specific implementation content of the method for recommending questions in a referenceable embodiment of the present invention has been described in detail in the above-mentioned method for recommending questions, so the repeated content will not be described here.

[0130] Figure 4 It is a schematic diagram of the main process of the method for recommending questions according to another referenceable embodiment of the present invention. As another embodiment of the present invention, the method for recommending questions may include the following steps:

[0131] Step 401: Perform intention recognition on the session information input by the user to obtain multiple fuzzy intentions.

[0132] Step 402: Determine whether the multiple fuzzy intentions exist in the node relationship graph; if so, execute Step 403; if not, end.

[0133] Step 403: Match the node questions corresponding to each fuzzy intention respectively according to the node relationship graph.

[0134] Step 404: Calculate the similarity between each node question and the session information input by the user respectively, and screen out the target node question with the highest similarity.

[0135] Step 405: Use the node where the target node question is located as the departure node, and match at least one destination node according to the node relationship graph.

[0136] Step 406: Sort the at least one destination node based on the relationship attributes between the departure node and the destination node, screen out at least one target destination node, and use the questions in the at least one target destination node as recommended questions.

[0137] Step 407: Push the at least one recommended question to the user.

[0138] Since the user's intention is fuzzy, it is impossible to update the node relationship graph based on the user's click results.

[0139] In addition, the specific implementation content of the method for recommending questions in another referenceable embodiment of the present invention has been described in detail in the above-mentioned method for recommending questions, so the repeated content will not be described herein again.

[0140] Figure 5 It is a schematic diagram of the main process of the method for recommending questions according to still another referenceable embodiment of the present invention. As another embodiment of the present invention, if the intent recognition result is an accurate intent and the user clicks on a certain recommended question, the steps of updating the node relationship graph may include:

[0141] Step 501, determine whether the accurate intent exists in the node relationship graph; if so, execute step 502; if not, execute step 503.

[0142] Step 502, use the node where the accurate intent is located as the starting node.

[0143] Step 503, create a new starting node, and use the accurate intent and its corresponding question (standard question method) as the attributes of the starting node.

[0144] Step 504, obtain the clicked recommended question from the click result of the user on the at least one recommended question.

[0145] Step 505, determine whether the clicked recommended question exists in the node relationship graph; if so, execute step 506; if not, execute step 507.

[0146] Step 506, use the node where the clicked recommended question is located as the destination node.

[0147] Step 507, create a new destination node, and use the clicked recommended question as the attribute of the destination node.

[0148] Step 508, update the relationship attribute between the starting node and the destination node.

[0149] Optionally, step 508 may include: obtaining all destination nodes having an associated relationship with the starting node, calculating the sum of the heat values between the starting node and all the destination nodes; calculating the heat percentage between the starting node and the destination node according to the heat value between the starting node and the destination node and the sum of the heat values between the starting node and all the destination nodes; determining the heat coefficient between the starting node and the destination node according to the number of relationship directions of the destination node; calculating the weight between the starting node and the destination node according to the heat percentage and the heat coefficient between the starting node and the destination node.

[0150] In an embodiment of the present invention, the weight between the departure node and the destination node can be calculated according to the heat percentage and the heat coefficient between the departure node and the destination node.

[0151] Optionally, the following formula can be used to calculate the heat percentage between the departure node and the destination node:

[0152] Heat percentage = Heat value between the departure node and the destination node / Sum of heat values between the departure node and all destination nodes.

[0153] Among them, the heat value can be the click frequency of the user. The higher the heat value, the more times of jumps between these two nodes, and the more concerned the user is.

[0154] Optionally, the following formula can be used to calculate the heat coefficient between the departure node and the destination node:

[0155] Heat coefficient = log10(Relationship count where the destination node is pointed to + 1)

[0156] Optionally, the following formula can be used to calculate the weight between the departure node and the destination node:

[0157] Weight = Heat percentage * Heat coefficient

[0158] For example:

[0159]

[0160] In addition, the specific implementation content of the method for recommending questions in another referenceable embodiment of the present invention has been described in detail in the above-mentioned method for recommending questions, so the repeated content will not be described here.

[0161] Figure 6 It is a schematic diagram of the main process of the method for recommending questions according to another referenceable embodiment of the present invention. As another embodiment of the present invention, if the intent recognition result is an accurate intent, the user's click result is none, and the user expresses the current intent, the steps of updating the node relationship graph may include:

[0162] Step 601, determine whether the accurate intent exists in the node relationship graph; if so, execute step 602; if not, execute step 603.

[0163] Step 602, use the node where the accurate intent is located as the departure node.

[0164] Step 603, create a new departure node, and use the accurate intent and its corresponding question (standard question method) as the attributes of the departure node.

[0165] Step 604: Determine whether the current intent exists in the node relationship graph. If yes, execute Step 608; if not, execute Step 605.

[0166] Step 605: Determine whether there is a problem (standard question) corresponding to the current intent in the node relationship graph. If yes, execute Step 606; if not, execute Step 607.

[0167] Step 606: Take the node where the problem corresponding to the current intent is located as the destination node.

[0168] Step 607: Create a new destination node and take the current intent as the attribute of the destination node.

[0169] Step 608: Take the node where the current intent is located as the destination node;

[0170] Step 609: Update the relationship attribute between the starting node and the destination node.

[0171] In addition, the specific implementation content of the method for recommending questions in another referenceable embodiment of the present invention has been described in detail in the above-mentioned method for recommending questions, so the repeated content will not be described here again.

[0172] Figure 7 It is a schematic diagram of the main modules of a device for recommending questions according to an embodiment of the present invention. As Figure 7 shown, the device 700 for recommending questions includes an identification module 701, a matching module 702, and a recommendation module 703. Among them, the identification module 701 is used to perform intent recognition on the session information input by the user to obtain an intent recognition result; the matching module 702 is used to match at least one recommended question according to the intent recognition result and the node relationship graph; wherein, the attribute of each node includes an intent and a question; the recommendation module 703 is used to push the at least one recommended question to the user.

[0173] Optionally, the identification module 701 is further used for:

[0174] Perform intent recognition on the session information input by the user to obtain multiple fuzzy intents or one accurate intent.

[0175] Optionally, if the intent recognition result is multiple fuzzy intents;

[0176] Then the matching module 702 is further used for:

[0177] Match the node questions corresponding to each fuzzy intent respectively according to the node relationship graph;

[0178] Calculate the similarity between each node problem and the session information respectively, and filter out the target node problem with the highest similarity;

[0179] Match at least one recommended problem according to the target node problem and the node relationship graph.

[0180] Optionally, the matching module 702 is further configured to:

[0181] Take the node where the target node problem is located as the starting node, and match at least one destination node according to the node relationship graph;

[0182] Sort the at least one destination node based on the relationship attributes between the starting node and the destination node, filter out at least one target destination node, and use the problems in the at least one target destination node as recommended problems.

[0183] Optionally, if the intention recognition result is an accurate intention;

[0184] Then the matching module 702 is further configured to:

[0185] Take the node where the accurate intention is located as the starting node, and match at least one destination node according to the node relationship graph;

[0186] Sort the at least one destination node based on the relationship attributes between the starting node and the destination node, filter out at least one target destination node, and use the problems in the at least one target destination node as recommended problems.

[0187] Optionally, the relationship attributes between the starting node and the destination node include weight value and / or heat value.

[0188] Optionally, it further includes an update module, which is used to:

[0189] After recommending the at least one recommended problem to the user, update the node relationship graph according to the click result of the user on the at least one recommended problem and the accurate intention.

[0190] Optionally, the update module is further configured to:

[0191] Obtain the clicked recommended problem from the click result of the user on the at least one recommended problem;

[0192] Take the node where the clicked recommended problem is located as the destination node, and take the node where the accurate intention is located as the starting node, and update the relationship attributes between the starting node and the destination node.

[0193] Optionally, the update module is further configured to:

[0194] If the click result of the user is none and the user expresses the current intention, then determine whether the current intention exists in the node relationship graph;

[0195] If not, create a new destination node, and the attributes of the destination node include the current intention and its corresponding question;

[0196] If so, use the node where the current intention is located as the destination node;

[0197] Use the node where the accurate intention is located as the starting node, and establish the relationship between the starting node and the destination node;

[0198] Update the relationship attributes between the starting node and the destination node.

[0199] Optionally, the update module is further configured to:

[0200] Obtain all destination nodes having an associated relationship with the starting node, and calculate the sum of the heat values between the starting node and all the destination nodes;

[0201] Calculate the heat percentage between the starting node and the destination node according to the heat value between the starting node and the destination node, and the sum of the heat values between the starting node and all the destination nodes;

[0202] Determine the heat coefficient between the starting node and the destination node according to the number of relationships pointed to by the destination node;

[0203] Calculate the weight between the starting node and the destination node according to the heat percentage and the heat coefficient between the starting node and the destination node.

[0204] According to the various embodiments described above, it can be seen that the present invention solves the technical problem of manually maintaining a recommended question list in the prior art by means of intention recognition of the session information input by the user to obtain an intention recognition result, and matching at least one recommended question according to the intention recognition result and the node relationship graph. The embodiment of the present invention matches at least one recommended question through the node relationship graph, thereby pushing the recommended question to the user, and automatically dynamically updating the node relationship graph according to the user's click result and input content, which can not only improve the accuracy of subsequent matching, but also does not require the operation and maintenance personnel to manually maintain the recommended question list. The embodiment of the present invention can improve the operation and maintenance efficiency, save 3% of the operation human efficiency, and also improve the user experience and increase the problem-solving rate by 5%.

[0205] It should be noted that the specific implementation content of the apparatus for recommending questions in the present invention has been described in detail in the method for recommending questions described above, so the repeated content will not be described herein again.

[0206] Figure 8 FIG. 800 shows an exemplary system architecture to which the method for recommending questions or the apparatus for recommending questions according to an embodiment of the present invention can be applied.

[0207] As Figure 8 shown, the system architecture 800 may include terminal devices 801, 802, 803, a network 804, and a server 805. The network 804 is used as a medium to provide a communication link between the terminal devices 801, 802, 803 and the server 805. The network 804 may include various connection types, such as wired, wireless communication links, or fiber optic cables, etc.

[0208] Users can use the terminal devices 801, 802, 803 to interact with the server 805 through the network 804 to receive or send messages, etc. Various communication client applications may be installed on the terminal devices 801, 802, 803, such as shopping applications, web browser applications, search applications, instant messaging tools, email clients, social platform software, etc. (only as examples).

[0209] The terminal devices 801, 802, 803 may be various electronic devices having a display screen and supporting web browsing, including but not limited to smart phones, tablet computers, laptop portable computers, and desktop computers, etc.

[0210] The server 805 may be a server providing various services, such as a background management server (only as an example) that supports shopping websites browsed by users using the terminal devices 801, 802, 803. The background management server may analyze and process data such as item information query requests received, and feedback the processing results (such as target push information, item information - only as examples) to the terminal devices.

[0211] It should be noted that the method for recommending questions provided by the embodiments of the present invention is generally executed by the server 805. Correspondingly, the apparatus for recommending questions is generally provided in the server 805.

[0212] It should be understood that Figure 8 the numbers of terminal devices, networks, and servers in

[0213] are merely illustrative. According to actual needs, there may be any number of terminal devices, networks, and servers. Figure 9 Figure 9 Figure 9The terminal device shown is only an example and should not impose any limitations on the functions and scope of use of the embodiments of the present invention.

[0214] As Figure 9 shown, the computer system 900 includes a central processing unit (CPU) 901, which can perform various appropriate actions and processes according to the program stored in the read-only memory (ROM) 902 or the program loaded from the storage section 908 into the random access memory (RAM) 903. In the RAM 903, various programs and data required for the operation of the system 900 are also stored. The CPU 901, ROM 902, and RAM 903 are connected to each other via a bus 904. The input / output (I / O) interface 905 is also connected to the bus 904.

[0215] The following components are connected to the I / O interface 905: an input section 906 including a keyboard, a mouse, etc.; an output section 907 including a cathode ray tube (CRT), a liquid crystal display (LCD), etc. and a speaker, etc.; a storage section 908 including a hard disk, etc.; and a communication section 909 including a network interface card such as a LAN card, a modem, etc. The communication section 909 performs communication processing via a network such as the Internet. A drive 910 is also connected to the I / O interface 905 as required. A removable medium 911, such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, etc., is installed on the drive 910 as required so that the computer program read from it can be installed into the storage section 908 as required.

[0216] Specifically, according to the embodiments disclosed in the present invention, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, the embodiments disclosed in the present invention include a computer program, which includes a computer program carried on a computer-readable medium, and the computer program contains program codes for performing the methods shown in the flowcharts. In such an embodiment, the computer program can be downloaded and installed from the network via the communication section 909, and / or installed from the removable medium 911. When the computer program is executed by the central processing unit (CPU) 901, the above functions defined in the system of the present invention are executed.

[0217] It should be noted that the computer-readable medium shown in the present invention can be a computer-readable signal medium, a computer-readable storage medium, or any combination of the two. A computer-readable storage medium can be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination of the above. More specific examples of a computer-readable storage medium can include, but are not limited to: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In the present invention, a computer-readable storage medium can be any tangible medium that contains or stores a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. In the present invention, a computer-readable signal medium can include a data signal propagated in a baseband or as part of a carrier wave, which carries computer-readable program code. Such a propagated data signal can take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. A computer-readable signal medium can also be any computer-readable medium other than a computer-readable storage medium, which can send, propagate, or transmit a program for use by or in conjunction with an instruction execution system, apparatus, or device. The program code contained on a computer-readable medium can be transmitted using any appropriate medium, including but not limited to: wireless, wire, optical fiber, RF, etc., or any suitable combination of the above.

[0218] The flowcharts and block diagrams in the accompanying drawings illustrate the possible architectures, functions, and operations of systems, methods, and computer programs according to various embodiments of the present invention. In this regard, each block in a flowchart or block diagram can represent a module, a program segment, or a part of code that contains one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions marked in the blocks can occur in a different order than that marked in the accompanying drawings. For example, two consecutive blocks shown can actually be executed substantially in parallel, and they can sometimes be executed in the reverse order, depending on the functions involved. It should also be noted that each block in the block diagram or flowchart, as well as the combination of blocks in the block diagram or flowchart, can be implemented by a dedicated hardware-based system for performing the specified functions or operations, or can be implemented by a combination of dedicated hardware and computer instructions.

[0219] The modules involved in the embodiments of the present invention can be implemented in software or in hardware. The described modules can also be provided in a processor. For example, it can be described as: a processor includes an identification module, a matching module, and a recommendation module. In some cases, the names of these modules do not constitute a limitation on the modules themselves.

[0220] As another aspect, the present invention further provides a computer-readable medium. The computer-readable medium can be included in the device described in the above embodiments; or it can exist alone without being assembled into the device. The above computer-readable medium carries one or more programs. When the one or more programs are executed by the device, the device is caused to include: performing intent recognition on the session information input by the user to obtain an intent recognition result; matching at least one recommended question according to the intent recognition result and the node relationship graph; where the attribute of each node includes an intent and a question; and pushing the at least one recommended question to the user.

[0221] According to the technical solution of the embodiments of the present invention, by adopting the technical means of performing intent recognition on the session information input by the user to obtain an intent recognition result and matching at least one recommended question according to the intent recognition result and the node relationship graph, the technical problem of manually maintaining a list of recommended questions in the prior art is overcome. The embodiments of the present invention match at least one recommended question through the node relationship graph, thereby pushing the recommended question to the user, and automatically dynamically updating the node relationship graph according to the user's click result and input content, which can not only improve the accuracy of subsequent matching, but also does not require the operation and maintenance personnel to manually maintain the list of recommended questions. The embodiments of the present invention can improve the operation and maintenance efficiency, save 3% of the operation human efficiency, and also improve the user experience and increase the problem-solving rate by 5%.

[0222] The above specific embodiments do not constitute a limitation on the protection scope of the present invention. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can occur depending on design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. A method for recommending problems, characterized in that, Including: Performing intent recognition on the session information input by the user to obtain an intent recognition result; According to the intent recognition result and the node relationship graph, matching at least one recommended question; wherein, the attributes of each node include an intent and a question, the node relationships are all one-way relationships, each relationship has two nodes, and the two nodes are a starting node and a destination node, and the relationship attributes between the starting node and the destination node include a weight value and / or a popularity value; Pushing the at least one recommended question to the user; If the intent recognition result is multiple fuzzy intents, then according to the intent recognition result and the node relationship graph, matching at least one recommended question, including: Respectively matching the node questions corresponding to each fuzzy intent according to the node relationship graph; Respectively calculating the similarity between each node question and the session information, and screening out the target node question with the highest similarity; According to the target node question and the node relationship graph, matching at least one recommended question.

2. The method according to claim 1, wherein Performing intent recognition on the session information input by the user to obtain an intent recognition result, including: Performing intent recognition on the session information input by the user to obtain multiple fuzzy intents or one accurate intent.

3. The method according to claim 1, characterized in that, According to the target node question and the node relationship graph, matching at least one recommended question, including: Taking the node where the target node question is located as the starting node, and matching at least one destination node according to the node relationship graph; Sorting the at least one destination node based on the relationship attributes between the starting node and the destination node, screening out at least one target destination node, and taking the questions in the at least one target destination node as recommended questions.

4. The method according to claim 2, wherein If the intent recognition result is one accurate intent; Then according to the intent recognition result and the node relationship graph, matching at least one recommended question, including: Taking the node where the accurate intent is located as the starting node, and matching at least one destination node according to the node relationship graph; Sorting the at least one destination node based on the relationship attributes between the starting node and the destination node, screening out at least one target destination node, and taking the questions in the at least one target destination node as recommended questions.

5. The method according to claim 4, characterized in that, After recommending the at least one recommended question to the user, further including: Updating the node relationship graph according to the click result of the user on the at least one recommended question and the accurate intent.

6. The method according to claim 5, characterized in that, Updating the node relationship graph according to the click result of the user on the at least one recommended question and the accurate intent, including: Obtaining the clicked recommended question from the click result of the user on the at least one recommended question; Taking the node where the clicked recommended question is located as the destination node, and taking the node where the accurate intent is located as the starting node, and updating the relationship attributes between the starting node and the destination node.

7. The method according to claim 5, characterized in that Updating the node relationship graph according to the click result of the user on the at least one recommended question and the accurate intent, including: If the user's click result is none and the user expresses the current intent, then determining whether the current intent exists in the node relationship graph; If not, create a new destination node, and the attributes of the destination node include the current intention and its corresponding question; If so, use the node where the current intention is located as the destination node; Use the node where the accurate intention is located as the starting node, and establish the relationship between the starting node and the destination node; Update the relationship attributes between the starting node and the destination node.

8. The method according to claim 6 or 7, characterized in that Updating the relationship attributes between the starting node and the destination node includes: Obtain all destination nodes having an associated relationship with the starting node, and calculate the sum of the heat values between the starting node and all the destination nodes; Calculate the heat percentage between the starting node and the destination node according to the heat value between the starting node and the destination node, and the sum of the heat values between the starting node and all the destination nodes; Determine the heat coefficient between the starting node and the destination node according to the number of relationships pointed to by the destination node; Calculate the weight between the starting node and the destination node according to the heat percentage and the heat coefficient between the starting node and the destination node.

9. A device for recommending questions, characterized in that, Including: An identification module, configured to perform intention identification on the session information input by the user to obtain an intention identification result; A matching module, configured to match at least one recommended question according to the intention identification result and the node relationship graph; wherein, the attributes of each node include an intention and a question, the node relationships are all one-way relationships, each relationship has two nodes, and the two nodes are the starting node and the destination node, and the relationship attributes between the starting node and the destination node include weights and / or heat values; A recommendation module, configured to push the at least one recommended question to the user; If the intention identification result is multiple fuzzy intentions, the matching module is further configured to: Match the node questions corresponding to each fuzzy intention according to the node relationship graph respectively; Calculate the similarity between each node question and the session information respectively, and filter out the target node question with the highest similarity; Match at least one recommended question according to the target node question and the node relationship graph.

10. An electronic device, characterized in that, Including: One or more processors; A storage device, configured to store one or more programs, When the one or more programs are executed by the one or more processors, the one or more processors implement the method according to any one of claims 1-8.

11. A computer-readable medium having a computer program stored thereon, characterized in that, The program, when executed by the processor, implements the method according to any one of claims 1-8.

Images