Method, device, apparatus and computer-readable storage medium for extracting subject matter

Abstract

The present invention relates to the field of financial technology, and discloses a method, device, equipment and computer-readable storage medium for extracting a topic. The method for extracting a topic includes: obtaining search log information and each search statement; determining a session in the search log information according to each search statement, wherein each search statement in the session is semantically associated; determining the topic distribution of the session through a topic model, and determining the uniform resource locator corresponding to the session according to the topic distribution; and extracting a target topic from each topic corresponding to the session according to the uniform resource locator corresponding to the session. The present invention improves the accuracy of topic extraction.

Description

Technical Field

[0001] The present invention relates to the technical field of financial technology (Fintech), and in particular to a subject extraction method, device, equipment and computer-readable storage medium. Background Art

[0002] With the development of computer technology, more and more technologies are being applied in the financial field. The traditional financial industry is gradually transforming into financial technology (Fintech). However, due to the security and real-time requirements of the financial industry, higher requirements are also placed on technology.

[0003] The implicit topics in the search log information can be used for the functions of the search engine. For example, researchers apply implicit topics to functions such as spelling correction and search personalization in the search engine. In other words, the search log information is a very valuable information library for improving the search results of the search engine.

[0004] However, existing solutions for mining implicit topics perform a global search for implicit topics in the entire log information, but some topics in the log information are not relevant, resulting in inaccurate topic extraction. Summary of the invention

[0005] The main purpose of the present invention is to provide a method, device, equipment and computer-readable storage medium for extracting a topic, aiming to solve the problem of inaccurate topic extraction.

[0006] To achieve the above object, the present invention provides a method for extracting a theme, the method comprising:

[0007] Get search log information and individual search statements;

[0008] Determining a session in the search log information according to each of the search statements, wherein each of the search statements in the session is semantically associated;

[0009] Determine a topic distribution of the conversation through a topic model, and determine a uniform resource locator corresponding to the conversation according to the topic distribution;

[0010] According to the uniform resource locator corresponding to the conversation, a target topic is extracted from each topic corresponding to the conversation.

[0011] Optionally, the step of determining a session in the search log information according to each of the search statements includes:

[0012] In the search log information, determining reference parameters corresponding to adjacent search statements, the reference parameters including at least one of keywords of the adjacent search statements and semantic similarity of the adjacent search statements;

[0013] According to the adjacent search statements corresponding to the reference parameters that meet the semantic condition, a session is extracted from the search log information, wherein the search statements in the session are sequentially adjacent and semantically related.

[0014] Optionally, the step of extracting a session from the search log information according to adjacent search statements corresponding to the reference parameters that satisfy the semantic condition includes:

[0015] In the search log information, determining the interval length between the search time points of adjacent search statements;

[0016] According to the interval duration and adjacent search statements corresponding to the reference parameters that meet the semantic conditions, a session is extracted from the search log information, wherein the search statements in the session are adjacent to each other and semantically related, and the interval duration corresponding to adjacent search statements in the session is less than a preset duration.

[0017] Optionally, the preset condition includes at least one of the following:

[0018] The semantic similarity of the adjacent search statements is greater than a preset similarity;

[0019] The keywords in adjacent search sentences are the same.

[0020] Optionally, the step of determining a uniform resource locator corresponding to the session according to the topic distribution includes:

[0021] Determining the marginal probability corresponding to each topic in the conversation according to the topic distribution;

[0022] A uniform resource locator corresponding to the session is determined according to the marginal probability corresponding to the session.

[0023] Optionally, before the step of determining the marginal probability corresponding to each topic in the conversation according to the topic distribution, the step further includes:

[0024] Determining whether the session is associated with a click operation record;

[0025] When determining the click operation record associated with the session, the step of determining the marginal probability corresponding to each topic in the session according to the topic distribution is performed.

[0026] Optionally, before the step of determining the topic distribution of the conversation by using a topic model, the step further includes:

[0027] Acquire each document information, wherein the document information includes tags, and the tags include a topic distribution tag, a word distribution tag, and a uniform resource locator distribution tag of the document information;

[0028] Inputting each of the document information into a preset model to train the preset model;

[0029] When the convergence value of the preset model is less than the preset convergence value, the training of the preset model is stopped, and the preset model whose training is stopped is saved as the topic model.

[0030] Optionally, the step of obtaining search log information and each search statement includes:

[0031] Obtain search log information, and determine the search engine corresponding to the search log information;

[0032] A search record of the search engine is obtained, and the search statement is determined according to the search record.

[0033] Optionally, the step of extracting a target topic from each topic corresponding to the conversation according to the uniform resource locator corresponding to the conversation includes:

[0034] According to the uniform resource locator corresponding to the conversation, parameter inference is performed on the conversation to obtain the probability of each topic in the conversation;

[0035] A topic with a maximum probability in the session is determined as a target topic, and the target topic is extracted from the session.

[0036] Optionally, when there are multiple search log information extraction sessions, parameter inference of each session is processed in parallel.

[0037] To achieve the above object, the present invention further provides a subject extraction device, the subject extraction device comprising:

[0038] The acquisition module is used to obtain search log information and each search statement;

[0039] A determination module, configured to determine a session in the search log information according to each of the search statements, wherein each of the search statements in the session is semantically associated;

[0040] The determination module is further configured to determine the topic distribution of the conversation through a topic model, and determine the uniform resource locator corresponding to the conversation according to the topic distribution;

[0041] The extraction module is used to extract a target topic from each topic corresponding to the conversation according to the uniform resource locator corresponding to the conversation.

[0042] To achieve the above-mentioned purpose, the present invention also provides a topic extraction device, which includes: a topic model, a memory, a processor, and an extraction program stored in the memory and executable on the processor, the topic model is connected to the processor, and the extraction program, when executed by the processor, implements the steps of the topic extraction method as described above.

[0043] To achieve the above object, the present invention further provides a computer-readable storage medium, on which an extraction program is stored, and when the extraction program is executed by a processor, the steps of the extraction method of the subject described above are implemented.

[0044] The present invention provides a method, device, equipment and computer-readable storage medium for extracting a topic. The present invention splits the search log information into sessions of search statements with semantic associations, and determines the topic distribution of the session through a topic model, and then determines the uniform resource locator of the session based on the topic distribution to accurately locate each topic in the session, and finally accurately extracts the target topic from each topic of the session. Compared with the current global search and extraction of implicit topics for the entire log information, the present invention can group semantically associated search statements into one session, so that each topic in the session is related, and then any topic extracted from the session can represent the implicit topic information of the session, overcoming the defect of the prior art that the topic information in the log information is irrelevant, resulting in inaccurate topic extraction, and improving the accuracy of topic extraction. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Figure 1 A schematic diagram of the structure of a device in a hardware operating environment involved in an embodiment of the present invention;

[0046] Figure 2 A schematic diagram of a first embodiment of the extraction method of the present invention;

[0047] Figure 3 This is a functional module diagram of the first embodiment of the extraction device of the subject of the present invention.

[0048] The realization of the purpose, functional features and advantages of the present invention will be further explained in conjunction with embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0049] It should be understood that the specific embodiments described herein are only used to explain the present invention, and are not used to limit the present invention.

[0050] Reference Figure 1 , Figure 1 The figure is a schematic diagram of the device structure of the hardware operating environment involved in the embodiment of the present invention.

[0051] The device involved in the embodiment of the present invention is an extraction device for connecting the subject.

[0052] like Figure 1 As shown, the extraction device of the connection theme may include: a processor 1001, such as a CPU, a communication bus 1002, a user interface 1003, a network interface 1004, and a memory 1005. Among them, the communication bus 1002 is used to realize the connection and communication between these components. The user interface 1003 may include a display screen (Display), an input unit such as a keyboard (Keyboard), and the optional user interface 1003 may also include a standard wired interface and a wireless interface. The network interface 1004 may optionally include a standard wired interface and a wireless interface (such as a Wi-Fi interface). The memory 1005 may be a high-speed RAM memory, or it may be a stable memory (non-volatile memory), such as a disk memory. The memory 1005 may also be a storage device independent of the aforementioned processor 1001.

[0053] Those skilled in the art will understand that Figure 1 The extraction device structure of the docking subject shown in the figure does not constitute a limitation of the extraction device of the docking subject, and may include more or less components than shown in the figure, or combine certain components, or arrange the components differently.

[0054] like Figure 1 As shown, the memory 1005 as a computer storage medium may include an operating system, a network communication module, a user interface module, and an extraction program.

[0055] exist Figure 1 In the terminal shown, the network interface 1004 is mainly used to connect to the server and perform data communication with the server; the user interface 1003 is mainly used to connect to the client and perform data communication with the client; and the processor 1001 can be used to call the extraction program stored in the memory 1005 and perform the following operations:

[0056] Get search log information and individual search statements;

[0057] Determining a session in the search log information according to each of the search statements, wherein each of the search statements in the session is semantically associated;

[0058] Determine a topic distribution of the conversation through a topic model, and determine a uniform resource locator corresponding to the conversation according to the topic distribution;

[0059] According to the uniform resource locator corresponding to the conversation, a target topic is extracted from each topic corresponding to the conversation.

[0060] In one embodiment, the processor 1001 may call the extraction program stored in the memory 1005, and further perform the following operations:

[0061] In the search log information, determining reference parameters corresponding to adjacent search statements, the reference parameters including at least one of keywords of the adjacent search statements and semantic similarity of the adjacent search statements;

[0062] According to the adjacent search statements corresponding to the reference parameters that meet the semantic condition, a session is extracted from the search log information, wherein the search statements in the session are sequentially adjacent and semantically related.

[0063] In one embodiment, the processor 1001 may call the extraction program stored in the memory 1005, and further perform the following operations:

[0064] In the search log information, determining the interval length between the search time points of adjacent search statements;

[0065] According to the interval duration and adjacent search statements corresponding to the reference parameters that meet the semantic conditions, a session is extracted from the search log information, wherein the search statements in the session are adjacent to each other and semantically related, and the interval duration corresponding to adjacent search statements in the session is less than a preset duration.

[0066] In one embodiment, the processor 1001 may call the extraction program stored in the memory 1005, and further perform the following operations:

[0067] The semantic similarity of the adjacent search statements is greater than a preset similarity;

[0068] The keywords in adjacent search sentences are the same.

[0069] In one embodiment, the processor 1001 may call the extraction program stored in the memory 1005, and further perform the following operations:

[0070] Determining the marginal probability corresponding to each topic in the conversation according to the topic distribution;

[0071] A uniform resource locator corresponding to the session is determined according to the marginal probability corresponding to the session.

[0072] In one embodiment, the processor 1001 may call the extraction program stored in the memory 1005, and further perform the following operations:

[0073] Determining whether the session is associated with a click operation record;

[0074] When determining the click operation record associated with the session, the step of determining the marginal probability corresponding to each topic in the session according to the topic distribution is performed.

[0075] In one embodiment, the processor 1001 may call the extraction program stored in the memory 1005, and further perform the following operations:

[0076] Acquire each document information, wherein the document information includes tags, and the tags include a topic distribution tag, a word distribution tag, and a uniform resource locator distribution tag of the document information;

[0077] Inputting each of the document information into a preset model to train the preset model;

[0078] When the convergence value of the preset model is less than the preset convergence value, the training of the preset model is stopped, and the preset model whose training is stopped is saved as the topic model.

[0079] In one embodiment, the processor 1001 may call the extraction program stored in the memory 1005, and further perform the following operations:

[0080] Obtain search log information, and determine the search engine corresponding to the search log information;

[0081] A search record of the search engine is obtained, and the search statement is determined according to the search record.

[0082] In one embodiment, the processor 1001 may call the extraction program stored in the memory 1005, and further perform the following operations:

[0083] According to the uniform resource locator corresponding to the conversation, parameter inference is performed on the conversation to obtain the probability of each topic in the conversation;

[0084] A topic with a maximum probability in the session is determined as a target topic, and the target topic is extracted from the session.

[0085] In one embodiment, the processor 1001 may call the extraction program stored in the memory 1005, and further perform the following operations:

[0086] When there are multiple search log information extraction sessions, parameter inference of each of the sessions is processed in parallel.

[0087] Based on the above hardware structure, various embodiments of the extraction method of the subject matter of the present invention are proposed.

[0088] Reference Figure 2 , Figure 2 The present invention is a flow chart of the first embodiment of the extraction method of the subject matter of the present invention, wherein the extraction method of the subject matter comprises:

[0089] Step S10, obtaining search log information and each search statement;

[0090] In this embodiment, the execution subject is the extraction device of the subject. For the convenience of description, the device is used below to refer to the extraction device of the subject. The device can collect search log information, which is a log generated by a user when using an application with a search engine. The search log information includes a search statement, which can be a statement entered by a user on the search interface of the application. Such statements are recorded by the application and form a search operation record, which is then sent to the device. The device associates the search operation record with the application and stores it. The search operation record only contains the search statement.

[0091] When the device needs to extract a topic, it obtains the search log information, determines the application that generates the search log information, that is, determines the search engine that generates the search log information, and then determines the search record corresponding to the search engine, thereby determining the search statement according to the search record. The search statement can be a sentence or a word.

[0092] Step S20, determining a session in the search log information according to each of the search statements, wherein each of the search statements in the session is semantically associated;

[0093] After determining the search statement, the device can determine the session in the search log information. Specifically, the search log information is distributed with various search statements, and each search statement has a corresponding position in the search log information, and the position reflects the adjacent relationship between the search statements.

[0094] The device first marks each search statement in the search log information. The device then determines the reference parameters of adjacent search statements based on the marks. The reference parameters include at least one of the semantic similarity between adjacent search statements and the semantic similarity of adjacent search statements. The device will determine whether the reference parameters of two adjacent search statements meet the semantic conditions. If the semantic conditions are met, it can be determined that the two adjacent search statements are in the same session, that is, if every two adjacent search statements meet the semantic conditions, the adjacent search statements are grouped into one session. It can be understood that the device extracts each session in the search log information based on the adjacent search statements corresponding to the reference parameters that meet the semantic conditions, and each search statement in the session is adjacent in sequence and needs to be semantically associated.

[0095] The semantic condition includes at least one of: the semantic similarity of the adjacent search statements is greater than the preset similarity and the keywords in the adjacent search statements are the same. A model for calculating semantic similarity is provided in the device. The device inputs the adjacent search statements into the model. The model performs semantic analysis on the adjacent search statements to calculate the semantic similarity of the adjacent search statements, that is, the device obtains the semantic similarity output by the model. The device can extract the keywords corresponding to the search statements. The keywords can be proper nouns in the search statements. For example, if the search statement is "query for high-speed rail to Chengdu", the keywords can be proper nouns "Chengdu" and "high-speed rail". The device can determine the keywords corresponding to each search statement. If the keywords of the adjacent search statements are the same, the adjacent search statements are semantically associated. Of course, if the search statement has 3 or more keywords, then the keywords that are the same in the adjacent search statements should be greater than one. At this time, the adjacent search statements can be identified as voice-associated. For example, search statement A has three keywords abc, and search statement B adjacent to search statement A has two keywords ab. Then search statement A and search statement B have two identical keywords ab, and it can be considered that the two are semantically associated.

[0096] The device determines in sequence whether the reference parameters of adjacent search statements meet the semantic conditions. If the reference parameters of the previous adjacent search statement meet the semantic conditions, and the reference parameters of the current adjacent search statement meet the semantic conditions, then the above three search statements belong to the same session. If the reference parameters of the previous adjacent search statement meet the semantic conditions, and the reference parameters of the current adjacent search statement do not meet the semantic conditions, the new search statement is divided into a new session. For example, if the previous adjacent search statement is AB, and the current adjacent search statement is BC, then C is a new search statement. The device can determine multiple sessions in this way, that is, the device extracts sessions from the search log information according to the characteristics that the search statements in the session are adjacent and semantically related in sequence. It should be noted that when determining the session, the interval duration of adjacent search statements also needs to be considered. If the interval duration is long, it can be determined that the two adjacent search statements belong to different sessions. In this regard, the device determines the interval duration between the search time points of the adjacent search statements, so as to extract each session from the search log information according to the interval duration and the adjacent search statements corresponding to the reference parameters of the fully loaded semantic conditions. Specifically, the search statements are associated with corresponding search time points, and the device can determine the interval duration between adjacent search statements based on the search time points. After obtaining the interval duration, the device determines whether the interval duration is less than the preset duration. The preset duration can be any composite value. For example, the preset duration can be 30 minutes. If the interval duration is less than the preset duration, the device obtains the reference parameters of the adjacent search statements, and determines whether the adjacent search statements are in the same session based on the reference parameters and semantic conditions. If the interval duration is greater than or equal to the preset duration, it can be directly determined that the two adjacent search statements are in different sessions. Of course, the device can first determine each session based on the reference parameters, and then determine the interval duration for the adjacent search statements in each session, thereby re-splitting a new session from the determined session.

[0097] Step S30, determining the topic distribution of the conversation through a topic model, and determining the uniform resource locator corresponding to the conversation according to the topic distribution;

[0098] The device is provided with a topic model, and the topic model can generate topics corresponding to each session. The topic model can be trained. Specifically, the device obtains multiple document information, and the document information includes tags, and the tags include topic distribution tags, word distribution tags, and uniform resource locator tags in the document information.

[0099] The device inputs the document information containing the label into the preset model to train the preset model. When the convergence value of the preset model is less than the preset convergence value, the training of the preset model can be stopped, and the preset model whose training is stopped is saved as the topic model.

[0100] The topic model first generates the topic distribution theta for each document, and then for each session, the device can determine the various topics of the session according to the topic distribution of the session. The device can determine the marginal probability P(z|theta) corresponding to each topic z in the session according to the topic distribution of the session. Specifically, the device determines the position of the topic in the session according to the topic distribution, and then randomly selects an marginal probability for each position, and the selected marginal probability is the marginal probability of the topic corresponding to the position. The topic distribution can be understood as the position of the topic in the session, so the device can directly determine the position of each topic in the session according to the topic distribution. The randomly selected marginal probability can be pre-set or generated in real time. The random selection method can be dice throwing, that is, each side or each point of the dice is associated with a preset marginal probability, and the marginal probability associated with the face or point presented after the dice is thrown is the marginal probability corresponding to the position.

[0101] After determining the marginal probabilities corresponding to each topic in the conversation, the device can determine the uniform resource locator corresponding to the conversation according to the marginal probabilities corresponding to the conversation. Specifically, a conversation generally has a user's click behavior, which can be understood as a click search action for a search statement in an input box. The click behavior is recorded in a click operation record, and the click operation record can be associated with the search statement, so the click operation record is associated with the conversation. The click operation record includes the user's click operation on the search statement. If the click operation record is associated with the conversation, a topic is randomly determined in the conversation, and the uniform resource locator P(URL|Z) is obtained according to the marginal probability of the topic.

[0102] Specifically, the device stores a plurality of uniform resource locators P and a plurality of corresponding relationships, and the corresponding relationship is the relationship between the face of the dice and the uniform resource locator P. When the device determines that a session is associated with a click operation record, it obtains the marginal probability of the randomly determined topic, and then determines the face presented after the dice is thrown by the marginal probability (the marginal probability is determined by the face presented after the dice is thrown, refer to the above description for details), and then extracts the corresponding relationship of the face, and finally obtains the uniform resource locator P by the corresponding relationship and the face presented after the dice is thrown, and the obtained uniform resource locator P is the uniform resource locator P of the session corresponding to the session.

[0103] It is understandable that when a session is associated with a click operation record, the marginal probability corresponding to each topic in the session is determined according to the topic distribution, and the uniform resource locator corresponding to the session is determined according to the marginal probability corresponding to the session. In this way, the device generates a uniform resource locator for the session with the click behavior.

[0104] Step S40: extracting a target topic from each topic corresponding to the conversation according to the uniform resource locator corresponding to the conversation.

[0105] After determining the uniform resource locator, the device can locate the position of each topic in the conversation according to the uniform resource locator, thereby extracting the target topic in the conversation. It should be noted that since the search statements in the conversation are semantically related, each topic can represent the implicit topic of the conversation, so the target topic can be any topic in the conversation.

[0106] In addition, the device can determine the topic with the highest probability as the target topic, and the topic with the highest probability is the implicit topic that is most appropriate to the conversation. In this regard, after calculating the uniform resource locator of the conversation, the device performs parameter inference on the conversation by means of variational inference to obtain the probability of each topic in the conversation, and then extracts the topic with the highest probability in each conversation as the target topic. In addition, the parameter inference process requires iterative calculation of multiple parameters, resulting in a large computational workload for parameter inference, and the search log information is generally divided into multiple sessions. The device will process the parameter inference of each session in parallel to reduce the workload. Specifically, the device uses the MapReduce computing framework to use multiple machines to perform variational inference in parallel. By using multiple machines to process training data in parallel, the training efficiency can be effectively enhanced, which also reduces the workload of parameter inference for multiple sessions.

[0107] In the technical solution provided in this embodiment, the device splits the search log information into sessions with semantically related search statements, and determines the topic distribution of the session through the topic model, and then determines the uniform resource locator of the session based on the topic distribution to accurately locate each topic in the session, and finally accurately extracts the target topic from each topic of the session. Compared with the current global search and extraction of implicit topics for the entire log information, the present invention can classify semantically related search statements into one session, so that each topic in the session is related, and then any topic extracted from the session can represent the implicit topic information of the session, overcoming the defect of the prior art that the topic information in the log information is not related, resulting in inaccurate topic extraction, and improving the accuracy of topic extraction.

[0108] The invention also provides a subject extraction device.

[0109] Reference Figure 3 , Figure 3 This is a functional module diagram of the first embodiment of the extraction device of the subject of the present invention.

[0110] like Figure 3 As shown, the extraction device of the subject comprises:

[0111] An acquisition module 10 is used to acquire search log information and each search statement;

[0112] A determination module 20, configured to determine a session in the search log information according to each of the search statements, wherein each of the search statements in the session is semantically associated;

[0113] The determination module 20 is further configured to determine the topic distribution of the conversation through a topic model, and determine the uniform resource locator corresponding to the conversation according to the topic distribution;

[0114] The extraction module 30 is used to extract a target topic from each topic corresponding to the conversation according to the uniform resource locator corresponding to the conversation.

[0115] In one embodiment, the subject extraction device comprises:

[0116] The determination module 20 is further configured to determine reference parameters corresponding to adjacent search statements in the search log information, wherein the reference parameters include at least one of keywords of adjacent search statements and semantic similarity of adjacent search statements;

[0117] The extraction module 30 is further configured to extract a session from the search log information according to adjacent search statements corresponding to the reference parameters that meet the semantic condition, wherein the search statements in the session are sequentially adjacent and semantically related.

[0118] In one embodiment, the subject extraction device comprises:

[0119] The determination module 20 is further configured to determine the interval between search time points of adjacent search statements in the search log information;

[0120] The extraction module 30 is further used to extract a session from the search log information according to the interval duration and adjacent search statements corresponding to the reference parameters that meet the semantic conditions, wherein the search statements in the session are adjacent and semantically related in sequence, and the interval duration corresponding to adjacent search statements in the session is less than a preset duration.

[0121] In one embodiment, the subject extraction device comprises:

[0122] The determination module 20 is further configured to determine the marginal probability corresponding to each topic in the conversation according to the topic distribution;

[0123] The determination module 20 is further configured to determine a uniform resource locator corresponding to the session according to the marginal probability corresponding to the session.

[0124] In one embodiment, the subject extraction device comprises:

[0125] A determination module, used to determine whether the session is associated with a click operation record;

[0126] An execution module is used to execute the step of determining the marginal probability corresponding to each topic in the session according to the topic distribution when determining the click operation record associated with the session.

[0127] In one embodiment, the subject extraction device further includes an acquisition module, an input module and a storage module:

[0128] An acquisition module, used to acquire each document information, wherein the document information includes tags, and the tags include a topic distribution tag, a word distribution tag, and a uniform resource locator distribution tag of the document information;

[0129] An input module, used for inputting each of the document information into a preset model to train the preset model;

[0130] The saving module is used to stop training the preset model when the convergence value of the preset model is less than the preset convergence value, and save the preset model whose training is stopped as a topic model.

[0131] In one embodiment, the subject extraction device further comprises:

[0132] An acquisition module, used to acquire search log information and determine the search engine corresponding to the search log information;

[0133] The acquisition module is used to acquire the search record of the search engine and determine the search statement according to the search record.

[0134] In one embodiment, the subject extraction device further includes an inference module:

[0135] An inference module, configured to perform parameter inference on the conversation according to the uniform resource locator corresponding to the conversation to obtain the probability of each topic in the conversation;

[0136] A determination module is used to determine the topic with the highest probability in the session as the target topic, and extract the target topic from the session.

[0137] In one embodiment, the subject extraction device further includes a processing module:

[0138] The processing module is used to perform parallel processing on parameter inference of each of the sessions when there are multiple search log information extraction sessions.

[0139] Among them, the functional implementation of each module in the extraction device of the above subject corresponds to the various steps in the extraction method embodiment of the above subject, and its functions and implementation processes will not be repeated here one by one.

[0140] The present invention also provides a computer-readable storage medium having an extraction program stored thereon, and when the extraction program is executed by a processor, the extraction method of the subject matter described in any one of the above embodiments is implemented.

[0141] The specific embodiments of the computer-readable storage medium of the present invention are basically the same as the embodiments of the extraction method of the above-mentioned subject matter, and will not be described in detail here.

[0142] It should be noted that, in this article, the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, article or system including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or system. In the absence of further restrictions, an element defined by the sentence "comprises a ..." does not exclude the existence of other identical elements in the process, method, article or system including the element.

[0143] The serial numbers of the above embodiments of the present invention are only for description and do not represent the advantages or disadvantages of the embodiments.

[0144] 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 a necessary general hardware platform, and of course by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present invention is essentially 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) as described above, and includes a number of instructions for a terminal device (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 invention.

[0145] The above are only preferred embodiments of the present invention, and are not intended to limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made using the contents of the present invention specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. A method for extracting a topic, characterized in that: The method for extracting the subject includes: Get search log information and individual search statements; Determine a session in the search log information according to each of the search statements, wherein each of the search statements in the session is semantically associated, and the session includes multiple topics; Determine a topic distribution of the conversation through a topic model, and determine a uniform resource locator corresponding to the conversation according to the topic distribution; Extracting a target topic from each topic corresponding to the conversation according to the uniform resource locator corresponding to the conversation, wherein each topic can represent an implicit topic of the conversation, and the target topic is any one of the topics; The step of determining the uniform resource locator corresponding to the session according to the topic distribution includes: Determine the position of each topic in the conversation according to the topic distribution, randomly select the marginal probability of the topic corresponding to each position by rolling a dice, and associate the face or number of points of the dice with the preset marginal probability; obtain the marginal probability of the randomly determined topic, and determine the face or number of points of the dice corresponding to the marginal probability of the topic; determine the uniform resource locator corresponding to the topic according to the corresponding relationship between the preset face or number of points of the dice and the uniform resource locator, and the uniform resource locator can locate the position of the topic in the conversation; The step of determining a session in the search log information according to each of the search statements includes: In the search log information, determining reference parameters corresponding to adjacent search statements, the reference parameters including at least one of keywords of the adjacent search statements and semantic similarity of the adjacent search statements; According to the adjacent search statements corresponding to the reference parameters that meet the semantic condition, a session is extracted from the search log information, wherein the search statements in the session are sequentially adjacent and semantically related.

2. The method for extracting a theme according to claim 1, characterized in that: The step of extracting a session from the search log information according to adjacent search statements corresponding to the reference parameters that meet the semantic condition comprises: In the search log information, determining the interval length between the search time points of adjacent search statements; According to the interval duration and adjacent search statements corresponding to the reference parameters that meet the semantic conditions, a session is extracted from the search log information, wherein the search statements in the session are adjacent to each other and semantically related, and the interval duration corresponding to adjacent search statements in the session is less than a preset duration.

3. The method for extracting a theme according to claim 1, characterized in that: The semantic condition includes at least one of the following: The semantic similarity of the adjacent search statements is greater than a preset similarity; The keywords in adjacent search sentences are the same.

4. The method for extracting a theme according to claim 1, wherein: Before the step of determining the marginal probability corresponding to each topic in the conversation according to the topic distribution, the method further includes: Determining whether the session is associated with a click operation record; When determining the click operation record associated with the session, the step of determining the marginal probability corresponding to each topic in the session according to the topic distribution is performed.

5. The method for extracting a theme according to claim 1, wherein: Before the step of determining the topic distribution of the conversation by using the topic model, the method further includes: Acquire each document information, wherein the document information includes tags, and the tags include a topic distribution tag, a word distribution tag, and a uniform resource locator distribution tag of the document information; Inputting each of the document information into a preset model to train the preset model; When the convergence value of the preset model is less than the preset convergence value, the training of the preset model is stopped, and the preset model whose training is stopped is saved as the topic model.

6. The method for extracting a theme according to claim 1, wherein: The step of obtaining search log information and each search statement includes: Obtain search log information, and determine the search engine corresponding to the search log information; A search record of the search engine is obtained, and the search statement is determined according to the search record.

7. The method for extracting the subject according to any one of claims 1 to 6, characterized in that: The step of extracting a target topic from each topic corresponding to the conversation according to the uniform resource locator corresponding to the conversation comprises: According to the uniform resource locator corresponding to the conversation, parameter inference is performed on the conversation to obtain the probability of each topic in the conversation; A topic with a maximum probability in the session is determined as a target topic, and the target topic is extracted from the session.

8. The method for extracting a theme according to claim 7, characterized in that: The performing parameter inference on the conversation to obtain the probability of each topic in the conversation includes: When there are multiple search log information extraction sessions, the parameter inference of each session is processed in parallel to obtain the probability of each topic in the session.

9. A theme extraction device, characterized in that: The extraction device of the subject comprises: The acquisition module is used to obtain search log information and each search statement; A determination module, configured to determine a session in the search log information according to each of the search statements, wherein each of the search statements in the session is semantically associated and the session includes a plurality of topics; The determination module is further configured to determine the topic distribution of the conversation through a topic model, and determine the uniform resource locator corresponding to the conversation according to the topic distribution; An extraction module, configured to extract a target topic from each topic corresponding to the conversation according to a uniform resource locator corresponding to the conversation, wherein each topic can represent an implicit topic of the conversation, and the target topic is any one of the topics; The determination module is further used to: determine the position of each topic in the conversation according to the topic distribution and randomly determine the marginal probability of the topic corresponding to each position by rolling a dice, wherein the face or number of points of the dice is associated with a preset marginal probability; obtain the marginal probability of the randomly determined topic, and determine the face or number of points of the dice corresponding to the marginal probability of the topic; determine the uniform resource locator corresponding to the topic according to the correspondence between the preset face or number of points of the dice and the uniform resource locator, wherein the uniform resource locator can locate the position of the topic in the conversation; The determination module is also used to: determine reference parameters corresponding to adjacent search statements in the search log information, the reference parameters including keywords of adjacent search statements and at least one of semantic similarities of adjacent search statements; extract conversations from the search log information based on adjacent search statements corresponding to the reference parameters that meet semantic conditions, wherein the search statements in the conversations are adjacent in sequence and semantically related.

10. A subject extraction device, characterized in that: The topic extraction device includes: a topic model, a memory, a processor, and an extraction program stored in the memory and executable on the processor, wherein the topic model is connected to the processor, and when the extraction program is executed by the processor, the steps of the topic extraction method as described in any one of claims 1 to 8 are implemented.

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

Images