Method, device, electronic device and storage medium for obtaining problem information

Abstract

The present application provides a method, device, electronic device, and storage medium for obtaining question information, which are applied to the computer field. The method includes: obtaining first text information to be processed; extracting candidate question information that matches the first text information from a preset question-and-answer library; filtering the implicit knowledge information in the first text information according to the implicit knowledge graph to obtain second text information; respectively obtaining the first semantic feature of the first text information, the second semantic feature in the second text information, and the candidate semantic feature in the candidate question information; and using the candidate question information whose feature similarity with the first semantic feature and the second semantic feature meets the similarity requirement as the target question information. The present application can reduce the interference caused by the semantic redundancy of implicit knowledge information on question information matching, improve the accuracy of question information extraction, and make the determined question more consistent with the user's actual questioning intention.

Description

Technical Field

[0001] The present application relates to the field of computers, and in particular to a method, device, electronic device, and storage medium for obtaining problem information. Background Art

[0002] Currently, FAQ (Frequently Asked Questions) technology generally collects and organizes frequently asked questions and answers in business scenarios. For a user's question, the most similar question is selected from these pairs and the answer to that question is returned to the user. If the user is unsure, several questions are selected and asked to confirm further. If the most similar question is also different from the user's question, the user is informed that the answer is unknown. FAQ technology is widely used in business scenarios such as training, robot customer service, health consulting, and insurance consulting.

[0003] However, the core of FAQ is to calculate the semantic similarity between two sentences (the user's question and the pre-collected questions). Whether based on characters or deep semantic coding, the matching is usually performed on the sentence itself, failing to consider the additional implicit meaning in the sentence. This results in the identified question information being inconsistent with the user's actual intention, reducing the accuracy of the obtained question information. Summary of the Invention

[0004] The embodiments of the present application provide a method, device, electronic device and computer-readable storage medium for obtaining question information, which aims to solve as much as possible the technical problem that in the process of the FAQ system in the related technology identifying the questions asked by the user, the additional implicit meaning in the sentence cannot be taken into account, which causes the identified question information to be inconsistent with the user's actual intention, thereby reducing the accuracy of the obtained question information.

[0005] This embodiment of the present application provides a method for obtaining problem information, the method comprising:

[0006] Obtaining a first text message to be processed;

[0007] Extracting candidate question information matching the first text information from a preset question-answer library;

[0008] Filtering the implicit knowledge information in the first text information according to the implicit knowledge graph to obtain second text information;

[0009] respectively acquiring a first semantic feature of the first text information, a second semantic feature in the second text information, and a candidate semantic feature in the candidate question information;

[0010] Semantic similarities between the candidate semantic feature and the first semantic feature and the second semantic feature are calculated respectively, and candidate question information whose semantic similarity meets the similarity requirement is used as target question information.

[0011] Optionally, filtering the implicit knowledge information in the first text information according to the implicit knowledge graph to obtain the second text information includes:

[0012] performing word segmentation processing on the first text information to obtain word segmentation information of the first text information;

[0013] mining, from the word segmentation information, word segmentation information of associated implicit knowledge information in the implicit knowledge graph as implicit knowledge information;

[0014] The implicit knowledge information in the first text information is filtered to obtain the second text information.

[0015] Optionally, the implicit knowledge graph includes at least: association relationships between graph elements, and the word segmentation information includes at least: first word segmentation information and second word segmentation information;

[0016] The mining of the word segmentation information from the word segmentation information, where the word segmentation information has associated implicit knowledge information in the implicit knowledge graph as implicit knowledge information, includes:

[0017] Querying the implicit knowledge graph for a first graph element that matches the first word segmentation information;

[0018] Querying the implicit knowledge graph for a second graph element associated with the first graph element;

[0019] When the second graph element matches the second word segmentation information, the second word segmentation information is used as implicit knowledge information.

[0020] Optionally, the graph element includes at least: an entity element and an attribute element, and the first word segmentation information includes at least: first sub-word segmentation information and second sub-word segmentation information;

[0021] The step of searching the implicit knowledge graph for a first graph element that matches the first word segmentation information includes:

[0022] When an entity element matching the first sub-segmentation information and an attribute element matching the second sub-segmentation information are queried in the implicit knowledge graph, and there is an association relationship between the entity element and the attribute element, the attribute element and the entity element are combined to obtain a first graph element.

[0023] Optionally, the second graph element includes at least: at least one of an entity element and an attribute element;

[0024] When the second graph element matches the second word segmentation information, using the second word segmentation information as implicit knowledge information includes:

[0025] When the matching degree between the first word segmentation information and the attribute element is within a preset matching degree range, determining target entity elements associated with the attribute element;

[0026] Performing word segmentation on each attribute element associated with each target entity element to obtain attribute element word segmentation;

[0027] According to the word frequency of each attribute element segmentation and the inverse document frequency of each attribute element in the implicit knowledge graph, the importance of each attribute element segmentation is obtained;

[0028] The attribute element segmentation whose importance is greater than the importance threshold is used as the target attribute element segmentation;

[0029] When the target attribute element segmentation matches the second segmentation information, the second segmentation information is used as implicit knowledge information.

[0030] Optionally, after segmenting the attribute element whose importance is greater than the importance threshold as the target attribute element segmentation, the method further includes:

[0031] When the target attribute element segmentation does not match the second segmentation information, the entity element associated with the attribute element segmentation with the highest importance is used as the target attribute element, and the re-determined target attribute element is used to enter the step of segmenting the attribute elements associated with the target attribute elements to obtain attribute element segmentations.

[0032] Optionally, after performing word segmentation processing on the first text information to obtain word segmentation information of the first text information, the method further includes:

[0033] Remove preset stop words from the word segmentation information.

[0034] Optionally, respectively calculating semantic similarities between the candidate semantic feature and the first semantic feature and the second semantic feature, and taking candidate question information whose semantic similarity meets similarity requirements as target question information includes:

[0035] Obtaining a first feature similarity between the candidate semantic feature and the first semantic feature, and a second feature similarity between the candidate semantic feature and the second semantic feature;

[0036] Among the candidate question information whose second feature similarity is greater than or equal to the first feature similarity, the candidate question information with the highest second feature similarity is used as the target question information.

[0037] Optionally, obtaining the first text information to be processed includes:

[0038] Receive the initial text message sent by the client;

[0039] The symbol information in the initial text information is removed to obtain the first text information.

[0040] Optionally, extracting candidate question information matching the first text information from a preset question-and-answer database includes:

[0041] Obtaining a matching degree between each preset question information in a preset text library and the first text information;

[0042] The preset question information whose matching degree meets the matching degree requirement is used as candidate question information.

[0043] Optionally, the taking the preset question information whose matching degree meets the matching degree requirement as candidate question information includes:

[0044] Calculating the order of matching values ​​of each of the preset question information;

[0045] The preset number of preset question information before the matching degree value order is used as candidate question information.

[0046] Optionally, respectively acquiring the first semantic feature of the first text information, the second semantic feature of the second text information, and the candidate semantic features of the candidate question information includes:

[0047] Inputting the first text information into a pre-trained deep semantic model to obtain a first semantic feature, and inputting the second text information into a pre-trained deep semantic model to obtain a second semantic feature;

[0048] The candidate semantic features are pre-extracted from the candidate question information through the deep semantic model.

[0049] Optionally, after filtering the implicit knowledge information in the first text information according to the implicit knowledge graph to obtain the second text information, the method further includes:

[0050] Filtering the implicit knowledge information in the candidate question information according to the implicit knowledge graph to obtain third text information;

[0051] respectively acquiring a second semantic feature in the second text information and a third semantic feature in the third text information;

[0052] The candidate question information whose feature similarity between the second semantic feature and the third semantic feature meets the similarity requirement is used as the target question information.

[0053] Optionally, the method further includes:

[0054] Extract the answer information corresponding to the target question information from the preset question and answer library.

[0055] The present invention provides a device for obtaining problem information, the device comprising:

[0056] A receiving module configured to obtain a first text message to be processed;

[0057] a processing module configured to extract candidate question information matching the first text information from a preset question-answer library;

[0058] Filtering the implicit knowledge information in the first text information according to the implicit knowledge graph to obtain second text information;

[0059] respectively acquiring a first semantic feature of the first text information, a second semantic feature in the second text information, and a candidate semantic feature in the candidate question information;

[0060] The output module is configured to respectively calculate the semantic similarity between the candidate semantic feature and the first semantic feature and the second semantic feature, and use the candidate question information whose semantic similarity meets the similarity requirement as the target question information.

[0061] Optionally, the processing module is further configured to:

[0062] performing word segmentation processing on the first text information to obtain word segmentation information of the first text information;

[0063] mining, from the word segmentation information, word segmentation information of associated implicit knowledge information in the implicit knowledge graph as implicit knowledge information;

[0064] The implicit knowledge information in the first text information is filtered to obtain the second text information.

[0065] Optionally, the implicit knowledge graph includes at least: association relationships between graph elements, and the word segmentation information includes at least: first word segmentation information and second word segmentation information;

[0066] Optionally, the processing module is further configured to:

[0067] Querying the implicit knowledge graph for a first graph element that matches the first word segmentation information;

[0068] Querying the implicit knowledge graph for a second graph element associated with the first graph element;

[0069] When the second graph element matches the second word segmentation information, the second word segmentation information is used as implicit knowledge information.

[0070] Optionally, the graph element includes at least: an entity element and an attribute element, and the first word segmentation information includes at least: first sub-word segmentation information and second sub-word segmentation information;

[0071] Optionally, the processing module is further configured to:

[0072] When an entity element matching the first sub-segmentation information and an attribute element matching the second sub-segmentation information are queried in the implicit knowledge graph, and there is an association relationship between the entity element and the attribute element, the attribute element and the entity element are combined to obtain a first graph element.

[0073] Optionally, the second graph element includes at least: at least one of an entity element and an attribute element;

[0074] Optionally, the processing module is further configured to:

[0075] When the matching degree between the first word segmentation information and the attribute element is within a preset matching degree range, determining target entity elements associated with the attribute element;

[0076] Performing word segmentation on each attribute element associated with each target entity element to obtain attribute element word segmentation;

[0077] According to the word frequency of each attribute element segmentation and the inverse document frequency of each attribute element in the implicit knowledge graph, the importance of each attribute element segmentation is obtained;

[0078] The attribute element segmentation whose importance is greater than the importance threshold is used as the target attribute element segmentation;

[0079] When the target attribute element segmentation matches the second segmentation information, the second segmentation information is used as implicit knowledge information.

[0080] Optionally, the processing module is further configured to:

[0081] When the target attribute element segmentation does not match the second segmentation information, the entity element associated with the attribute element segmentation with the highest importance is used as the target attribute element, and the re-determined target attribute element is used to enter the step of segmenting the attribute elements associated with the target attribute elements to obtain attribute element segmentations.

[0082] Optionally, the processing module is further configured to:

[0083] Remove preset stop words from the word segmentation information.

[0084] Optionally, the output module is further configured to:

[0085] Obtaining a first feature similarity between the candidate semantic feature and the first semantic feature, and a second feature similarity between the candidate semantic feature and the second semantic feature;

[0086] Among the candidate question information whose second feature similarity is greater than or equal to the first feature similarity, the candidate question information with the highest second feature similarity is used as the target question information.

[0087] Optionally, the receiving module is further configured to:

[0088] Receive the initial text message sent by the client;

[0089] The symbol information in the initial text information is removed to obtain the first text information.

[0090] Optionally, the processing module is further configured to:

[0091] Obtaining a matching degree between each preset question information in a preset text library and the first text information;

[0092] The preset question information whose matching degree meets the matching degree requirement is used as candidate question information.

[0093] Optionally, the processing module is further configured to:

[0094] Calculating the order of matching values ​​of each of the preset question information;

[0095] The preset number of preset question information before the matching degree value order is used as candidate question information.

[0096] Optionally, the processing module is further configured to:

[0097] Inputting the first text information into a pre-trained deep semantic model to obtain a first semantic feature, and inputting the second text information into a pre-trained deep semantic model to obtain a second semantic feature;

[0098] The candidate semantic features are pre-extracted from the candidate question information through the deep semantic model.

[0099] Optionally, the processing module is further configured to:

[0100] Filtering the implicit knowledge information in the candidate question information according to the implicit knowledge graph to obtain third text information;

[0101] respectively acquiring a second semantic feature in the second text information and a third semantic feature in the third text information;

[0102] The candidate question information whose feature similarity between the second semantic feature and the third semantic feature meets the similarity requirement is used as the target question information.

[0103] Optionally, the output module is further configured to:

[0104] Extract the answer information corresponding to the target question information from the preset question and answer library.

[0105] An embodiment of the present application provides an electronic device, comprising: a processor, a memory, and a computer program stored in the memory and executable on the processor, wherein the computer program, when executed by the processor, implements the steps of the method for obtaining problem information described above.

[0106] An embodiment of the present application provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the steps of the method for obtaining question information as described above are implemented.

[0107] The embodiments of the present application provide a method, device, electronic device, and computer-readable storage medium for obtaining question information. The methods mine implicit knowledge information in text information through an implicit knowledge graph, filter the implicit information in text sentences, and perform similarity comparison with candidate question information in a question-and-answer knowledge base by eliminating semantic features before and after the implicit knowledge information. This reduces the interference of semantic redundancy of implicit knowledge information on question information matching, improves the accuracy of question information extraction, and makes the determined question more consistent with the user's actual questioning intention. BRIEF DESCRIPTION OF THE DRAWINGS

[0108] Figure 1 A flowchart schematically illustrates a method for obtaining question information in an embodiment of the present application;

[0109] Figure 2 A schematic diagram schematically illustrates a relationship diagram of an implicit knowledge graph in an embodiment of the present application;

[0110] Figure 3 One of the flowcharts schematically illustrates another method for obtaining question information in an embodiment of the present application;

[0111] Figure 4 The second flowchart schematically illustrates another method for obtaining question information provided in an embodiment of the present application;

[0112] Figure 5 Schematically illustrating a third step flow chart of another method for obtaining question information provided in an embodiment of the present application;

[0113] Figure 6 A schematic diagram schematically illustrates the principle of an implicit knowledge graph in an embodiment of the present application;

[0114] Figure 7 Schematically illustrating a fourth step flow chart of another method for obtaining question information provided in an embodiment of the present application;

[0115] Figure 8 Schematically illustrating another relationship diagram of an implicit knowledge graph provided in an embodiment of the present application;

[0116] Figure 9 Schematically illustrating a fifth step flow chart of another method for obtaining question information provided in an embodiment of the present application;

[0117] Figure 10 Schematically illustrating a sixth step flow chart of another method for obtaining question information provided in an embodiment of the present application;

[0118] Figure 11 Schematically illustrating the seventh step of a flowchart of another method for obtaining question information provided in an embodiment of the present application;

[0119] Figure 12 Schematically illustrating the eighth step flow chart of another method for obtaining question information provided in an embodiment of the present application;

[0120] Figure 13 The following schematically illustrates a structural diagram of a device for obtaining question information in an embodiment of the present application. DETAILED DESCRIPTION

[0121] The following will be combined with the drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are part of the embodiments of this application, not all of them. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0122] In related technologies, FAQs typically pre-collect and organize frequently asked questions and answers (Q&A) in business scenarios. For a user's question, the most similar question is selected from these pairs and the answer to that question is returned to the user. If the user is unsure, several questions are selected for further confirmation. If the most similar question is also different from the user's question, the user is informed that the answer is unknown. FAQ technology is widely used in business scenarios such as training, robot customer service, health consulting, and insurance consulting.

[0123] The core of FAQs is to calculate the semantic similarity between two sentences (the user's question and the pre-collected questions). Whether based on characters or deep semantic encoding, matching is usually performed on the sentence itself, rarely using additional knowledge. This can lead to matching errors in some cases, such as when implicit knowledge is present in the sentence. Taking "What should I do if I have stye in my eye?" as an example, we searched for the question "What should I do if I have stye in my eye?" in an existing FAQ system and sorted it by similarity (calculated based on the trained BERT model) to obtain the following candidates in Table 1:

[0124]

[0125]

[0126] Table 1

[0127] The above ranking is incorrect. The first and second candidate questions have different meanings from the user's question, while the third sentence ("What should I do if I get a stye") is the same. However, because the third sentence does not contain the word "eye," its similarity during matching is not as high as the first two, resulting in a lower ranking. In fact, there is implicit knowledge between "eye" and "stye": stye occurs in the eye, so stye implies information about the eye. In other words, "eye" can actually be omitted in this question. If the word "eye" is removed from the user's question, changing it to "'What should I do if I get a stye'," the similarity with the candidate question "What should I do if I get a stye" will be greatly improved, and the ranking will move up, and the ranking result may be correct.

[0128] The use of implicit knowledge is not uncommon in question-answering scenarios. Table 2 lists several search statements and the implicit knowledge they contain:

[0129]

[0130] Table 2

[0131] If the words with implicit relationships in the question are processed before matching, the matching effect of similar questions can be effectively improved, thereby improving the accuracy of question answering. Based on this, this paper proposes a method for obtaining question information, aiming to improve the accuracy of the process of obtaining question information containing implicit knowledge.

[0132] Figure 1 A flowchart schematically illustrates a method for obtaining question information provided in an embodiment of the present application, the method comprising:

[0133] Step 101: Obtain a first text message to be processed.

[0134] It should be noted that the execution subject of the embodiment of the present application can be installed on a terminal device or server, which has data processing, data transmission and other functions to obtain text information and extract question information therein. Specifically, the embodiment of the present application can be used in a terminal device or server in an FAQ question-answering system to identify and extract question information from text information converted from text information submitted by users, and to reply with corresponding answer information.

[0135] In an embodiment of the present application, after the user submits initial information in text or voice form to the server through an application client with a question or consultation function, the server can extract text from the initial information to obtain the first text information that needs to be identified as question information.

[0136] Step 102: extract candidate question information matching the first text information from a preset question and answer database.

[0137] It should be noted that the preset question and answer database is a database that pre-stores the relationship between question information and answer information, wherein associated answer information is set for different question information, so that the system can automatically extract the corresponding answer information based on different question information and automatically answer user questions.

[0138] In an embodiment of the present application, considering that the form of the question information in the preset question library is fixed, while the ways in which users ask questions are diverse, similarity matching is performed directly in the preset question and answer library based on the first text information. The system may query multiple candidate question information with a similarity threshold greater than, for example, 80% in the preset question and answer library, and it is necessary to further screen out the question information that best meets the user's actual question intention from these candidate question information.

[0139] For example, assuming that the first text message is "What should I do if I have a stye?", the following Table 3 can be found in the preset question library:

[0140] Sorting Candidate question information 1 What to do if you have pimples on your eyes 2 What to do if blisters appear on your eyes 3 Causes of styes on the eyelids 4 What to do if you have stye 5 What to do if your child gets stye 6 What causes pinholes? 7 What should I do if there is a bag on my eyelid? 8 Causes of fat particles under the eyes 9 How to treat eye blisters 10 What to do if fat particles grow in the corners of the eyes

[0141] Table 3

[0142] It can be seen that there are 10 candidate question information in Table 3, but only the fourth item "What should I do if I get a stye" is a question information that meets the user's actual question intention. Therefore, the following will further explain how to filter out the correct question information from the candidate question information.

[0143] Step 103: Filter the implicit knowledge information in the first text information according to the implicit knowledge graph to obtain the second text information.

[0144] It should be noted that an implicit knowledge graph refers to a knowledge graph that stores the association relationships between element nodes, where different element nodes can represent various entities and relationship attributes, and these entity nodes are connected by edges representing the relationships between them. Implicit knowledge information refers to the segmentation to which the association relationship points in the second text information, among at least two segmentations that exist and store the association relationship in the implicit knowledge graph. For example, the association relationship between "needle eye" and "eye" is that "needle eye" points to "eye", so the implicit knowledge information in the first text information of "what to do if you have a stye in your eye" is "eye" instead of "needle eye". It can be understood that since there is actual semantic overlap between the words in the association relationship in the implicit knowledge graph, but the words pointed to by the association relationship are generally more superordinate, it will interfere with the matching of the question information. It is necessary to filter out the implicit knowledge information to eliminate this interference.

[0145] In an embodiment of the present application, the question-answering system can mine the implicit knowledge information in each word segment of the first text information based on the entity / attribute relationship stored in the implicit knowledge graph, and further filter the implicit knowledge information in the first text information to obtain the second text information for subsequent further processing.

[0146] For example, by performing word segmentation on the candidate question information in Table 3 above, the following results can be obtained in Table 4:

[0147] Questions Word segmentation results What to do if you have stye in your eyes What to do if you have a stye in your eyes? What to do if you have pimples on your eyes What to do if I have a pimple on my eye What to do if blisters appear on your eyes What to do if you have blisters on your eyes Causes of styes on the eyelids Eyelid / Style / Causes What to do if you have stye What to do if you have a stye? What to do if your child gets stye What to do if a child gets a stye What causes pinholes? How is the stye caused? What should I do if there is a bag on my eyelid? What to do if I have a bag on my eyelids Causes of fat particles under the eyes Causes of fat particles growing under the eyes How to treat eye blisters How to treat eye blisters What to do if fat particles grow in the corners of the eyes What to do if fat particles grow in the corners of the eyes

[0148] Table 4

[0149] Then the word segmentation results in each candidate question information are combined with the existing implicit relationships and queried in the implicit knowledge graph. The partial relationship structure diagram of the "needle eye" node in the graph can be seen in Figure 2 The mining results for the query "needle eye" are shown in Table 5:

[0150]

[0151] Table 5

[0152] Therefore, by filtering “eye” from “what to do if you have stye in your eyes”, you can get the second text information “what to do if you have stye in your eyes”.

[0153] Step 104 : respectively obtain the first semantic feature of the first text information, the second semantic feature of the second text information, and the candidate semantic features of the candidate question information.

[0154] It should be noted that the first semantic feature refers to a feature vector that can reflect the semantic features of each word in the first text information. The second semantic feature and the candidate semantic feature first semantic feature are the same and will not be described in detail here.

[0155] In an embodiment of the present application, the question-answering system may use a pre-trained semantic extraction model for extraction. The semantic extraction model may be an encoder with a representational feature extraction function, or other machine models with a feature extraction function, so as to obtain semantic features in each text information and question-answering information for subsequent similarity comparison. There is no limitation on this here.

[0156] For example, semantic features are extracted from the candidate question information, the first text information, and the second text information in Tables 3, 4, and 5 above to obtain the following Table 6:

[0157]

[0158] Table 6

[0159] Step 105 : Calculate semantic similarities between the candidate semantic feature and the first semantic feature and the second semantic feature, and use candidate question information whose semantic similarity meets similarity requirements as target question information.

[0160] In the embodiment of the present application, considering that the second text information after the implicit knowledge information in the past may not be more capable of representing the user's actual question intention, it is necessary to compare the similarity of the first semantic feature and the second semantic feature before and after the implicit knowledge information is removed with the candidate semantic feature, so as to screen out the target question information closest to the user's actual question intention based on the similarity. The candidate question information with the greatest similarity between the second semantic feature and the candidate semantic feature can be used as the target question information, or the candidate question information with the greatest similarity between the first semantic feature, the second semantic feature and the candidate semantic feature can be used as the target question information. The specific similarity requirement can be set according to actual needs and is not limited here.

[0161] For example, similarity calculation is performed on the semantic features in Table 6 to obtain the results shown in Table 7 below:

[0162] Sorting Questions (without removing implicit knowledge information) Similarity Questions (removing implicit knowledge information) Similarity 1 What to do if blisters appear on your eyes 0.985 What to do if you have stye 0.991 2 What to do if you have pimples on your eyes 0.984 What to do if your child gets stye 0.985 3 What to do if you have stye 0.974 What to do if you have pimples on your eyes 0.963 4 What to do if your child gets stye 0.959 What to do if blisters appear on your eyes 0.96 5 What to do if fat particles grow in the corners of the eyes 0.94 How to treat eye blisters 0.935 6 How to treat eye blisters 0.934 What to do if fat particles grow in the corners of the eyes 0.926 7 What should I do if there is a bag on my eyelid? 0.932 What should I do if there is a bag on my eyelid? 0.919 8 Causes of styes on the eyelids 0.875 Causes of styes on the eyelids 0.866 9 Causes of fat particles under the eyes 0.853 Causes of fat particles under the eyes 0.844 10 What causes pinholes? 0.796 What causes pinholes? 0.817

[0163] Table 7

[0164] Observations show that the accuracy of the semantic similarity of the second text information after the implicit knowledge information is removed is significantly improved compared to the original one. The correct candidate question information "What should I do if I get a stye" with the same semantics as the first text information has risen from the original third place to the first place and is higher than the highest similarity between the first text information and the candidate question information before the implicit knowledge information is removed. The correct matching of the candidate question information and the first text information can express the user's true intention of asking the question.

[0165] In the embodiment of the present application, implicit knowledge information in text information is mined through an implicit knowledge graph, and the implicit information in the text sentences is filtered. By eliminating the semantic features before and after the implicit knowledge information and performing a similarity comparison with the candidate question information in the question-answering knowledge base, the interference caused by the semantic redundancy of the implicit knowledge information on the matching of question information is reduced, the accuracy of question information extraction is improved, and the determined question is more in line with the user's actual question intention.

[0166] Optionally, refer to Figure 3 , the step 102 may include:

[0167] Step 201: perform word segmentation processing on the first text information to obtain word segmentation information of the first text information.

[0168] In the embodiment of the present application, each word in the first text information can be divided by a preset word segmentation library. For details, please refer to the word segmentation method in the relevant technology, which will not be described here.

[0169] Step 202 , mining from the word segmentation information, word segmentation information of associated implicit knowledge information in the implicit knowledge graph as implicit knowledge information.

[0170] In an embodiment of the present application, each word segmentation information is input into the implicit knowledge graph for mining. If multiple implicit knowledge information are mined simultaneously, the multiple implicit knowledge information can be combined as the implicit knowledge information of the first text information.

[0171] Step 203: Filter the implicit knowledge information in the first text information to obtain second text information.

[0172] In an embodiment of the present application, if there are multiple implicit knowledge information in the question sentence of the first text information, the multiple implicit knowledge information are omitted respectively, encoded, and the similarity between them and the candidate question information is calculated respectively, and compared with the similarity results between the first text information and the candidate question information, and the candidate question information with the highest similarity is taken as the final result.

[0173] Specifically, refer to Figure 4 , where Q represents the first text information, k1 represents implicit knowledge information 1, and k2 represents implicit knowledge 2, so that Q-k1 without implicit knowledge information 1, Q-k2 without implicit knowledge information 2, and Q-k1-k2 without implicit knowledge information k1 and implicit knowledge information k2 can be semantically encoded respectively. After extracting semantic features, the semantic feature similarities between Q-k1, Q-k2, Q-k1-k2 and Q respectively, and the semantic feature similarities between Q-k1, Q-k2, Q-k1-k2 and candidate question information respectively are compared, so as to select the candidate question information with the highest similarity as the target question information for final recognition.

[0174] Optionally, the implicit knowledge graph includes at least: the association relationship between graph elements, the word segmentation information includes at least: first word segmentation information, second word segmentation information, Figure 5 , the step 202 may include:

[0175] It should be noted that graph elements are at least the elements represented by each node in the implicit knowledge graph. The element type can be noun entities, attribute information, time information, etc., and can be set according to actual needs and is not limited here. The first segmentation information and the second segmentation information refer to the segmentation in the first text information. The first segmentation information and the second segmentation information can also be further divided. This is only for the convenience of subsequent distinction and is not limited here.

[0176] Step 2021: Search the implicit knowledge graph for a first graph element that matches the first word segmentation information.

[0177] Step 2022: Query the implicit knowledge graph for a second graph element that has an association relationship with the first graph element.

[0178] Step 2023: When the second graph element matches the second word segmentation information, the second word segmentation information is used as implicit knowledge information.

[0179] In the embodiment of steps 2021 to 2023 of the present application, through the pre-relation list R = [r0, r1...rn] in the implicit knowledge graph, where rn represents each graph element, the segmentation information w in the segmentation information list W = [w0, w1, w2...wn] is combined with the relationship r in the relationship list to perform a query in the implicit knowledge graph, and a list of results that meet the query is obtained as the second graph element. If the second graph element contains the segmentation information w' in the segmentation information list W, the segmentation combination w / w' is used as the implicit knowledge of the first text information, and w' is used as the implicit knowledge information to be passed.

[0180] Optionally, the graph element includes at least: an entity element and an attribute element, and the first word segmentation information includes at least: a first sub-word segmentation information and a second sub-word segmentation information. The step 2022 may include: when an entity element matching the first sub-word segmentation information and an attribute element matching the second sub-word segmentation information are searched in the implicit knowledge graph, and when there is an association relationship between the entity element and the attribute element, the attribute element and the entity element are combined to obtain a first graph element.

[0181] In an embodiment of the present application, the mining of implicit knowledge information can also be carried out by combining the word segmentation information of the candidate question information as entity elements and attribute elements to query the implicit knowledge graph. If there are entity elements and attribute elements in the implicit knowledge graph that are equal to and have a relationship with the word segmentation information, then the word segmentation combination w / w' is used as the implicit knowledge information of the question. The word segmentation information can also be combined to form a phrase as attribute information for query in the implicit knowledge graph.

[0182] For ease of understanding, refer to Figure 6 , where the entity elements and attribute elements in the knowledge graph are connected based on the triples of attribute relations, and the entity elements 1 and 2 are connected based on the triples of entity relations.

[0183] Optionally, refer to Figure 7 The step 2023 may include:

[0184] Step 20231: When the matching degree between the first word segmentation information and the attribute element is within a preset matching degree range, determine target entity elements associated with the attribute element.

[0185] Step 20232: Segment each attribute element associated with each target entity element to obtain attribute element segmentation.

[0186] Step 20233: Obtain the importance of each attribute element segmentation based on the word frequency of each attribute element segmentation and the inverse document frequency of each attribute element in the implicit knowledge graph.

[0187] Step 20234: Use the attribute element segmentation whose importance is greater than the importance threshold as the target attribute element segmentation.

[0188] Step 20235: When the target attribute element segmentation matches the second segmentation information, the second segmentation information is used as implicit knowledge information.

[0189] In the embodiment of steps 20231 to 20235 of the present application, for the case where the word segmentation information of the candidate question information exists in the attribute element and has high importance but is not completely equal to the attribute element, all attribute elements of the entity elements associated with the attribute element can be segmented to obtain the TF (TermFrequency) value of the attribute element word segmentation. At the same time, the IDF (InverseDocumentFrequency) value calculated by the attribute element word segmentation in the English-Chinese knowledge base is used to calculate the TF-IDF (importance) of the attribute element word segmentation, and it is used as an indicator to measure the importance of the attribute element word segmentation. The attribute element word segmentation with a TF-IDF value higher than the importance threshold is obtained. If the word segmentation information exists in the filtered attribute element word segmentation, it is determined that implicit knowledge information exists in the first text information, and the word combination is used as the implicit knowledge information of the first text information.

[0190] Step 20236: When the target attribute element segmentation does not match the second segmentation information, the entity element associated with the attribute element segmentation with the highest importance is used as the target attribute element, and the process proceeds to step 20232 using the newly determined target attribute element.

[0191] In an embodiment of the present application, the indirect implicit relationship between the word segmentation information can also be mined through the implicit knowledge graph. When the implicit knowledge information cannot be directly mined based on the first text information, after step 20234, after obtaining the attribute element word segmentation with a TF-IDF value higher than the set threshold, the entity element associated with the attribute element word segmentation corresponding to the maximum TF-IDF value is selected as the target entity element and the query is performed again by entering step 20232 to obtain the indirect implicit relationship.

[0192] For example, assuming that the first text information is "What should pregnant women eat if they have less amniotic fluid?", the structure of the indirect implicit relationship between pregnant women and amniotic fluid in the implicit knowledge graph is as follows: Figure 8 .

[0193] Among them, the word segmentation result of the first text information is 'pregnant women / amniotic fluid / eat / what'. When querying with 'amniotic fluid' as the entity element, no attribute element containing 'pregnant women' is found. At this time, it is necessary to refer to the TF-IDF value of the word segmentation of the attribute elements around the node, among which the value of 'fetus' is the largest, and the knowledge 'amniotic fluid / fetus' is obtained. Then, after querying with 'fetus' as the entity, the knowledge of 'pregnant women-pregnancy-fetus' is obtained, thereby obtaining the indirect implicit knowledge information 'pregnant women / amniotic fluid'. After the implicit knowledge 'pregnant women' is removed, 'what to eat with less amniotic fluid' is obtained. The subsequent similarity calculation and other steps are the same as above, and the most similar candidate is finally obtained as 'what to eat with less amniotic fluid'. To further verify the effectiveness of this method, this method is used to perform similarity matching on the example sentences in Table 2, and the calculation results of the similarity between the second text information and the correct candidate question information and the incorrect candidate question information are obtained, and compared with the calculation results corresponding to the first text information without removing the implicit knowledge information. The results are shown in Tables 8 and 9 below:

[0194]

[0195]

[0196] Table 8

[0197]

[0198] Table 9

[0199] The observation results show that after removing the implicit knowledge information in the first text information, the similarity between the second text information and the correct candidate similar question is improved. At the same time, its similarity with the incorrect candidate results is reduced, and the similarity between the correct similar question and the correct candidate is the highest among all candidates, which ultimately enables the retrieval to obtain a correct match.

[0200] Optionally, after step 201, the method may further include: removing preset stop words from the word segmentation information.

[0201] In the embodiment of the present application, stop words refer to words that are suspended by user settings or system default settings. The specific settings can be based on actual needs and are not limited here.

[0202] Optionally, refer to Figure 9 , the step 105 may include:

[0203] Step 1051 : Obtain a first feature similarity between the candidate semantic feature and the first semantic feature, and a second feature similarity between the candidate semantic feature and the second semantic feature.

[0204] Step 1052 : Among the candidate question information whose second feature similarity is greater than or equal to the first feature similarity, the candidate question information with the highest second feature similarity is used as the target question information.

[0205] In the embodiment of step 1051 to step 1052 of the present application, the cosine similarity between the candidate semantic feature and the first semantic feature and the second semantic feature can be calculated respectively to obtain the first feature similarity and the second feature pixel point. Considering that for the correct question information, the second feature similarity after removing the implicit knowledge information cannot be lower than the first feature similarity before removing the implicit knowledge information, when performing similarity screening, it is necessary to target the candidate question information with the highest similarity among the candidate semantic features whose first feature similarity is less than or equal to the second feature similarity as the final target question information, so as to avoid the situation where the extracted question information is inconsistent with the user's actual intention due to the removal of implicit knowledge information.

[0206] Optionally, step 101 may include: receiving an initial text message sent by a client, removing symbol information from the initial text message, and obtaining the first text message.

[0207] In an embodiment of the present application, the initial text information sent by the user through the client can be filtered for symbol information using the interference removal technology in the related art to avoid interference caused by the symbol information in extracting the problem information.

[0208] Optionally, refer to Figure 10 , the step 102 includes:

[0209] Step 1021: Obtain the matching degree between each preset question information in the preset text library and the first text information.

[0210] Step 1022: The preset question information whose matching degree meets the matching degree requirement is used as candidate question information.

[0211] In the embodiment of steps 1021 to 1022 of the present application, for a question-and-answer database with a smaller data volume, the first text information can be directly retrieved from the question-and-answer database using text retrieval (ES search engine / Annoy algorithm).

[0212] Optionally, refer to Figure 11 , the step 1022 may include:

[0213] Step 10221: Calculate the order of matching values ​​of each of the preset question information.

[0214] Step 10222: Use a preset number of preset question information before the matching degree value order as candidate question information.

[0215] In the embodiment of steps 10221 to 10222 of the present application, for a question-and-answer database with a larger data volume, a sorting algorithm (coarse sorting / fine sorting) can be used to sort the candidate question information, and finally obtain several candidate question information sets with high rankings.

[0216] Optionally, step 104 may include: inputting the first text information into a pre-trained deep semantic model to obtain a first semantic feature, and inputting the second text information into a pre-trained deep semantic model to obtain a second semantic feature, wherein the candidate semantic feature is pre-extracted from the candidate question information through the deep semantic model.

[0217] In this embodiment of the present application, the deep semantic model can be an encoder based on a deep semantic algorithm, pre-trained using question information from the question-answer database. Since the candidate question information is also pre-trained in the question-answer database, the candidate semantic features obtained by inputting the question information from the question-answer database into the deep semantic model are pre-stored and can be directly called upon when needed, eliminating the need to encode the candidate question information each time, thereby improving the efficiency of question information extraction.

[0218] Optionally, refer to Figure 12 After step 103, the method further includes:

[0219] Step 301: Filter the implicit knowledge information in the candidate question information according to the implicit knowledge graph to obtain third text information.

[0220] Step 302: Acquire a second semantic feature in the second text information and a third semantic feature in the third text information respectively.

[0221] Step 303 : The candidate question information whose feature similarity between the second semantic feature and the third semantic feature meets the similarity requirement is used as the target question information.

[0222] In the embodiment of step 301 to step 303 of the present application, it is also possible to mine implicit knowledge of the first text information and the candidate question information at the same time, filter out the implicit knowledge information in the candidate question information and the first text information, and then directly calculate the semantic feature similarity, without separately calculating the semantic similarity between the first text information before and after the removal of the implicit knowledge information, the second text information and the candidate question information. However, this method cannot directly use the question information encoding generated in advance by the model, and the candidate question information needs to be re-encoded each time a match is made. In terms of computing speed, for example Figure 1 In the embodiment of the method for obtaining question information, the method of only filtering the implicit knowledge of the first text information for mining is slower.

[0223] Optionally, after step 105 or step 303, the method further includes: extracting answer information corresponding to the target question information from a preset question and answer library.

[0224] In an embodiment of the present application, the preset question and answer library stores the association between question information and answer information. Therefore, when a candidate question information is determined to be the target question information, the system will query the preset question and answer library for the answer information corresponding to the target question information and send it to the user, so that the user can easily obtain the answer to the medical question he raised.

[0225] For example, assuming that the first text information is "What should I do if I get stye", the target question information finally determined by the method for obtaining question information in the above embodiment is "What should I do if I get stye", then the answer information corresponding to the target question information can be queried from the preset question and answer library, which is "The use of antibacterial drugs is the main treatment plan, and local anti-inflammatory lotions and eye ointments are required, such as levofloxacin lotion and tobramycin lotion. Eye ointments can choose gatifloxacin eye ointment or tobramycin eye ointment." Of course, this is just an exemplary description, and the association between specific question information and answer information can be set according to actual needs, and is not limited here.

[0226] In the embodiment of the present application, implicit knowledge information in text information is mined through an implicit knowledge graph, and the implicit information in the text sentences is filtered. By eliminating the semantic features before and after the implicit knowledge information and performing a similarity comparison with the candidate question information in the question-answering knowledge base, the interference caused by the semantic redundancy of the implicit knowledge information on the matching of question information is reduced, the accuracy of question information extraction is improved, and the determined question is more in line with the user's actual question intention.

[0227] Figure 13 The following is a schematic diagram of a structure of a device 40 for obtaining question information provided in an embodiment of the present application, wherein the device includes:

[0228] Receiving module 401, configured to obtain a first text message to be processed;

[0229] The processing module 402 is configured to extract candidate question information matching the first text information from a preset question and answer library;

[0230] Filtering the implicit knowledge information in the first text information according to the implicit knowledge graph to obtain second text information;

[0231] respectively acquiring a first semantic feature of the first text information, a second semantic feature in the second text information, and a candidate semantic feature in the candidate question information;

[0232] The output module 403 is configured to respectively calculate the semantic similarity between the candidate semantic feature and the first semantic feature and the second semantic feature, and use the candidate question information whose semantic similarity meets the similarity requirement as the target question information.

[0233] Optionally, the processing module 402 is further configured to:

[0234] performing word segmentation processing on the first text information to obtain word segmentation information of the first text information;

[0235] mining, from the word segmentation information, word segmentation information of associated implicit knowledge information in the implicit knowledge graph as implicit knowledge information;

[0236] The implicit knowledge information in the first text information is filtered to obtain the second text information.

[0237] Optionally, the implicit knowledge graph includes at least: association relationships between graph elements, and the word segmentation information includes at least: first word segmentation information and second word segmentation information;

[0238] Optionally, the processing module 402 is further configured to:

[0239] Querying the implicit knowledge graph for a first graph element that matches the first word segmentation information;

[0240] Querying the implicit knowledge graph for a second graph element associated with the first graph element;

[0241] When the second graph element matches the second word segmentation information, the second word segmentation information is used as implicit knowledge information.

[0242] Optionally, the graph element includes at least: an entity element and an attribute element, and the first word segmentation information includes at least: first sub-word segmentation information and second sub-word segmentation information;

[0243] Optionally, the processing module 402 is further configured to:

[0244] When an entity element matching the first sub-segmentation information and an attribute element matching the second sub-segmentation information are queried in the implicit knowledge graph, and there is an association relationship between the entity element and the attribute element, the attribute element and the entity element are combined to obtain a first graph element.

[0245] Optionally, the second graph element includes at least: at least one of an entity element and an attribute element;

[0246] Optionally, the processing module 402 is further configured to:

[0247] When the matching degree between the first word segmentation information and the attribute element is within a preset matching degree range, determining target entity elements associated with the attribute element;

[0248] Performing word segmentation on each attribute element associated with each target entity element to obtain attribute element word segmentation;

[0249] According to the word frequency of each attribute element segmentation and the inverse document frequency of each attribute element in the implicit knowledge graph, the importance of each attribute element segmentation is obtained;

[0250] The attribute element segmentation whose importance is greater than the importance threshold is used as the target attribute element segmentation;

[0251] When the target attribute element segmentation matches the second segmentation information, the second segmentation information is used as implicit knowledge information.

[0252] Optionally, the processing module 402 is further configured to:

[0253] When the target attribute element segmentation does not match the second segmentation information, the entity element associated with the attribute element segmentation with the highest importance is used as the target attribute element, and the re-determined target attribute element is used to enter the step of segmenting the attribute elements associated with the target attribute elements to obtain attribute element segmentations.

[0254] Optionally, the processing module 402 is further configured to:

[0255] Remove preset stop words from the word segmentation information.

[0256] Optionally, the output module 403 is further configured to:

[0257] Obtaining a first feature similarity between the candidate semantic feature and the first semantic feature, and a second feature similarity between the candidate semantic feature and the second semantic feature;

[0258] Among the candidate question information whose second feature similarity is greater than or equal to the first feature similarity, the candidate question information with the highest second feature similarity is used as the target question information.

[0259] Optionally, the receiving module 401 is further configured to:

[0260] Receive the initial text message sent by the client;

[0261] The symbol information in the initial text information is removed to obtain the first text information.

[0262] Optionally, the processing module 402 is further configured to:

[0263] Obtaining a matching degree between each preset question information in a preset text library and the first text information;

[0264] The preset question information whose matching degree meets the matching degree requirement is used as candidate question information.

[0265] Optionally, the processing module 402 is further configured to:

[0266] Calculating the order of matching values ​​of each of the preset question information;

[0267] The preset number of preset question information before the matching degree value order is used as candidate question information.

[0268] Optionally, the processing module 402 is further configured to:

[0269] Inputting the first text information into a pre-trained deep semantic model to obtain a first semantic feature, and inputting the second text information into a pre-trained deep semantic model to obtain a second semantic feature;

[0270] The candidate semantic features are pre-extracted from the candidate question information through the deep semantic model.

[0271] Optionally, the processing module 402 is further configured to:

[0272] Filtering the implicit knowledge information in the candidate question information according to the implicit knowledge graph to obtain third text information;

[0273] respectively acquiring a second semantic feature in the second text information and a third semantic feature in the third text information;

[0274] The candidate question information whose feature similarity between the second semantic feature and the third semantic feature meets the similarity requirement is used as the target question information.

[0275] Optionally, the output module 403 is further configured to:

[0276] Extract the answer information corresponding to the target question information from the preset question and answer library.

[0277] In the embodiment of the present application, implicit knowledge information in text information is mined through an implicit knowledge graph, and the implicit information in the text sentences is filtered. By eliminating the semantic features before and after the implicit knowledge information and performing a similarity comparison with the candidate question information in the question-answering knowledge base, the interference caused by the semantic redundancy of the implicit knowledge information on the matching of question information is reduced, the accuracy of question information extraction is improved, and the determined question is more in line with the user's actual question intention.

[0278] An embodiment of the present application provides an electronic device, comprising: a processor, a memory, and a computer program stored in the memory and executable on the processor, wherein the computer program, when executed by the processor, implements the steps of the method for obtaining problem information described above.

[0279] An embodiment of the present application provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the steps of the method for obtaining question information as described above are implemented.

[0280] Optionally, an embodiment of the present application also provides an electronic device, comprising: a processor, a memory, and a computer program stored in the memory and runnable on the processor. When the computer program is executed by the processor, the various processes of the above-mentioned information processing method embodiment are implemented and the same technical effect can be achieved. To avoid repetition, it will not be repeated here.

[0281] The present application also provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the various processes of the above-mentioned information processing method embodiment are implemented, and the same technical effects are achieved. To avoid repetition, the details are not described here. The computer-readable storage medium is, for example, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0282] It should be noted that, in this document, the terms "comprises," "includes," or any other variations thereof are intended to encompass non-exclusive inclusion, such that a process, method, article, or apparatus comprising a series of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, method, article, or apparatus. In the absence of further limitations, an element defined by the phrase "comprising a ..." does not exclude the presence of other identical elements in the process, method, article, or apparatus comprising the element.

[0283] Through the description of the above implementation methods, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus the necessary general hardware platform, and of course can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art, can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk), and includes a number of instructions for enabling a terminal (which can be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods described in each embodiment of the present application.

[0284] The embodiments of the present application are described above in conjunction with the accompanying drawings, but the present application is not limited to the above-mentioned specific implementation methods. The above-mentioned specific implementation methods are merely illustrative and not restrictive. Under the guidance of the present application, ordinary technicians in this field can also make many forms without departing from the scope of protection of the purpose of the present application and the claims, all of which are within the protection of the present application.

[0285] Those skilled in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed in the embodiments of this application can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0286] Those skilled in the art will clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0287] In the embodiments provided in this application, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0288] The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of these units may be selected to achieve the purpose of this embodiment according to actual needs.

[0289] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit.

[0290] If the functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes various media that can store program codes, such as a USB flash drive, a mobile hard disk, a ROM, a RAM, a magnetic disk, or an optical disk.

[0291] The above description is merely a specific embodiment of the present application, but the scope of protection of the present application is not limited thereto. Any changes or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in this application should be included in the scope of protection of the present application. Therefore, the scope of protection of the present application should be based on the scope of protection of the claims.

Claims

1. A method for obtaining problem information, characterized in that: The method comprises: Obtaining a first text message to be processed; Extracting candidate question information matching the first text information from a preset question-answer library; Filtering the implicit knowledge information in the first text information according to the implicit knowledge graph to obtain second text information; respectively acquiring a first semantic feature of the first text information, a second semantic feature in the second text information, and a candidate semantic feature in the candidate question information; Calculating semantic similarities between the candidate semantic feature and the first semantic feature and the second semantic feature respectively, and taking candidate question information whose semantic similarity meets the similarity requirement as target question information; The filtering of the implicit knowledge information in the first text information according to the implicit knowledge graph to obtain the second text information includes: Performing word segmentation processing on the first text information to obtain word segmentation information of the first text information; mining, from the word segmentation information, word segmentation information of associated implicit knowledge information in the implicit knowledge graph as implicit knowledge information; filtering the implicit knowledge information in the first text information to obtain second text information; The implicit knowledge graph includes the associated word segmentation information of semantic repetition.

2. The method according to claim 1, characterized in that The implicit knowledge graph at least includes: the association relationship between graph elements, and the word segmentation information at least includes: first word segmentation information and second word segmentation information; The mining of the word segmentation information from the word segmentation information, where the word segmentation information has associated implicit knowledge information in the implicit knowledge graph as implicit knowledge information, includes: Querying the implicit knowledge graph for a first graph element that matches the first word segmentation information; Querying the implicit knowledge graph for a second graph element associated with the first graph element; When the second graph element matches the second word segmentation information, the second word segmentation information is used as implicit knowledge information.

3. The method according to claim 2, characterized in that The graph elements include at least: entity elements and attribute elements, and the first segmentation information includes at least: first sub-segmentation information and second sub-segmentation information; The step of searching the implicit knowledge graph for a first graph element that matches the first word segmentation information includes: When an entity element matching the first sub-segmentation information and an attribute element matching the second sub-segmentation information are queried in the implicit knowledge graph, and there is an association relationship between the entity element and the attribute element, the attribute element and the entity element are combined to obtain a first graph element.

4. The method according to claim 2, characterized in that The second graph element includes at least: at least one of an entity element and an attribute element; When the second graph element matches the second word segmentation information, using the second word segmentation information as implicit knowledge information includes: When the matching degree between the first word segmentation information and the attribute element is within a preset matching degree range, determining target entity elements associated with the attribute element; Performing word segmentation on each attribute element associated with each target entity element to obtain attribute element word segmentation; According to the word frequency of each attribute element segmentation and the inverse document frequency of each attribute element in the implicit knowledge graph, the importance of each attribute element segmentation is obtained; The attribute element segmentation whose importance is greater than the importance threshold is used as the target attribute element segmentation; When the target attribute element segmentation matches the second segmentation information, the second segmentation information is used as implicit knowledge information.

5. The method according to claim 4, characterized in that After segmenting the attribute element whose importance is greater than the importance threshold as the target attribute element segmentation, the method further includes: When the target attribute element segmentation does not match the second segmentation information, the entity element associated with the attribute element segmentation with the highest importance is used as the target attribute element, and the re-determined target attribute element is used to enter the step of segmenting the attribute elements associated with the target attribute elements to obtain the attribute element segmentation.

6. The method according to claim 1, wherein The calculating the semantic similarities between the candidate semantic feature and the first semantic feature and the second semantic feature respectively, and taking the candidate question information whose semantic similarity meets the similarity requirement as the target question information, includes: Obtaining a first feature similarity between the candidate semantic feature and the first semantic feature, and a second feature similarity between the candidate semantic feature and the second semantic feature; Among the candidate question information whose second feature similarity is greater than or equal to the first feature similarity, the candidate question information with the highest second feature similarity is used as the target question information.

7. The method according to claim 1, characterized in that The step of respectively acquiring the first semantic feature of the first text information, the second semantic feature of the second text information, and the candidate semantic features of the candidate question information includes: Inputting the first text information into a pre-trained deep semantic model to obtain a first semantic feature, and inputting the second text information into a pre-trained deep semantic model to obtain a second semantic feature; The candidate semantic features are pre-extracted from the candidate question information through the deep semantic model.

8. The method according to claim 1, characterized in that After filtering the implicit knowledge information in the first text information according to the implicit knowledge graph to obtain the second text information, the method further includes: Filtering the implicit knowledge information in the candidate question information according to the implicit knowledge graph to obtain third text information; respectively acquiring a second semantic feature in the second text information and a third semantic feature in the third text information; The candidate question information whose feature similarity between the second semantic feature and the third semantic feature meets the similarity requirement is used as the target question information.

9. A device for obtaining problem information, characterized in that: The device comprises: A receiving module configured to obtain a first text message to be processed; a processing module configured to extract candidate question information matching the first text information from a preset question-answer library; Filtering the implicit knowledge information in the first text information according to the implicit knowledge graph to obtain second text information; respectively acquiring a first semantic feature of the first text information, a second semantic feature in the second text information, and a candidate semantic feature in the candidate question information; an output module configured to calculate semantic similarities between the candidate semantic feature and the first semantic feature and the second semantic feature, respectively, and use the candidate question information whose semantic similarity meets the similarity requirement as target question information; Wherein, the processing module is further configured to: Performing word segmentation processing on the first text information to obtain word segmentation information of the first text information; mining, from the word segmentation information, word segmentation information of associated implicit knowledge information in the implicit knowledge graph as implicit knowledge information; filtering the implicit knowledge information in the first text information to obtain second text information; The implicit knowledge graph includes the associated word segmentation information of semantic repetition.

10. An electronic device, characterized in that: include: A processor, a memory, and a computer program stored in the memory and executable on the processor, wherein the computer program, when executed by the processor, implements the steps of the method for obtaining question information according to any one of claims 1 to 8.

11. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps of the method for obtaining question information according to any one of claims 1 to 8.

Images