A reply information determination method and device

Abstract

The application discloses a reply information determination method and device, which is applied to the field of artificial intelligence technology and comprises the following steps: in response to a trigger instruction for a target area, target search information is displayed; a preset database corresponding to the target search information is acquired; based on a deep learning algorithm, a first candidate search information set matched with the target search information is determined from the preset database; based on a string clustering algorithm, a second candidate search information set matched with the target search information is determined from the first candidate search information set; in response to a trigger instruction for at least one second candidate search information in the second candidate search information set, target candidate information is determined; based on the preset database, target reply information corresponding to the target candidate information is determined; and the target reply information is taken as the reply information of the target search information. The application improves the accuracy of reply information in an artificial intelligence product and improves the information query experience of users.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of artificial intelligence, and in particular to a method and device for determining reply information. BACKGROUND

[0002] In the development process of AI intelligent products, the number and quality of question and answer pairs are crucial. Because in the experience process of users, questions that are in the middle of the rules (such as: "Ding-ding-ding, sing a song of XX", "Ding-ding-ding, how is the weather today") cannot completely satisfy the curiosity and desire of AI communication of users. Users often ask some special questions or knowledge questions (such as: "Ding-ding-ding, why is there a rainbow after the rain", "Ding-ding-ding, why does the height increase after getting up in the morning", "Ding-ding-ding, why was the Nibuchuo Treaty signed"). If such user questions cannot be answered, the AI intelligent product will appear to be not very smart, and the user will also lose interest in the product and no longer want to interact and experience it. As can be seen, the number and quality of question and answer pairs are crucial to AI intelligent products. With as many question and answer pairs as possible, the AI product will answer more user questions and bring more surprises to users.

[0003] In addition, the logic construction of question and answer pairs is the key to determining the degree of intelligence of a product. In the actual experience of users, all users have different ways of asking about the same question. For example, in the field of ten thousand whys, about the question of why penguins cannot fly, the user's possible ways of asking are as follows: "Why can't penguins fly" "Why can't penguins fly" "Why can't penguins fly" "Why can't penguins fly" …… It can be seen that users have different ways of asking about the same question, but they are all asking about the same thing. In addition, sometimes the same user will use different questions but the same intention to test the degree of intelligence of the AI product. The example is shown above. The user may change the order of words, replace synonyms, add some irrelevant words such as mood words and auxiliary words, etc. If the user asks: "Ding-ding-ding, why can't penguins fly" and gets an answer, and then asks: "Ding-ding-ding, why can't penguins fly" and gets another answer, it will seriously affect the user experience; if similar questions have the same theme but different answers, it will often make the user feel that the product is not smart enough, the data construction is not perfect enough, and the service is not stable enough. However, if different answers need to be configured for each similar question, it will also require a huge amount of manpower, material resources and financial resources.

[0004] Therefore, the application provides a reply information determination method and device, which sets the same reply information for search information with the same intention and theme in the process of constructing a question and answer pair, avoids setting different reply information for search information with the same intention and theme but different expression manners, and thus improves user experience. SUMMARY

[0005] The application provides a reply information determination method and device, which can improve the accuracy of reply information in artificial intelligence products and thus improves the information query experience of users.

[0006] In one aspect, the application provides a reply information determination method, which comprises:

[0007] displaying target search information in response to a trigger instruction for a target area;

[0008] obtaining a preset database corresponding to the target search information; the preset database stores the correspondence between search information and reply information;

[0009] determining a first candidate search information set matched with the target search information from the preset database based on a deep learning algorithm;

[0010] determining a second candidate search information set matched with the target search information from the first candidate search information set based on a string clustering algorithm;

[0011] determining target candidate information in response to a trigger instruction for at least one second candidate search information in the second candidate search information set;

[0012] determining target reply information corresponding to the target candidate information based on the preset database; and taking the target reply information as the reply information of the target search information.

[0013] In another aspect, a reply information determination device is provided, which comprises:

[0014] a target search information display module configured to display target search information in response to a trigger instruction for a target area;

[0015] a preset database obtaining module configured to obtain a preset database corresponding to the target search information; the preset database stores the correspondence between search information and reply information;

[0016] a first candidate search information set determination module configured to determine a first candidate search information set matched with the target search information from the preset database based on a deep learning algorithm;

[0017] determine a second candidate search information set matching the target search information from the first candidate search information set based on a string clustering algorithm;

[0018] a target candidate information determination module configured to determine target candidate information in response to a trigger instruction for at least one second candidate search information in the second candidate search information set;

[0019] a target reply information determination module configured to determine target reply information corresponding to the target candidate information based on the preset database, and take the target reply information as reply information of the target search information.

[0020] In another aspect, a reply information determination device is provided, which includes a processor and a memory, the memory storing at least one instruction or at least one program, the at least one instruction or the at least one program being loaded and executed by the processor to implement the reply information determination method as described above.

[0021] In another aspect, a computer storage medium is provided, which stores at least one instruction or at least one program, the at least one instruction or at least one program being loaded and executed by a processor to implement the reply information determination method as described above.

[0022] In another aspect, a computer program product or computer program is provided, which includes computer instructions stored in a computer readable storage medium. A processor of a computer device reads the computer instructions from the computer readable storage medium, and the processor executes the computer instructions to cause the computer device to perform to implement the reply information determination method as described above.

[0023] The reply information determination method and device provided by the present application have the following technical effects:

[0024] In the process of constructing the question and answer pair, when a target search information is input by a development user, a set of to-be-recommended search information (i.e., a second candidate search information set) similar to the theme and intent of the target search information is determined based on a deep learning algorithm and a string clustering algorithm in sequence. The development user can select a target candidate information identical to the theme and intent of the target search information based on the second candidate search information set. The similarity between the set of to-be-recommended search information and the target search information is improved, so that the development user can quickly determine the target candidate information identical to the theme and intent of the target search information. The target reply information corresponding to the target candidate information is taken as the reply information of the target search information, which effectively avoids setting different reply information for similar search information, improves the accuracy of the reply information, and improves the information query experience of the user. BRIEF DESCRIPTION OF DRAWINGS

[0025] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, and the advantages thereof, simple introductions to the drawings needed to be used in the embodiments or the prior art description will be given below. Obviously, the drawings in the following description are only some embodiments of the present application, and for those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0026] Figure 1 is a schematic diagram of a reply information determination system provided by an embodiment of the present application;

[0027] Figure 2 is a flowchart of a reply information determination method provided by an embodiment of the present application;

[0028] Figure 3 is a flowchart of a method for determining a first candidate search information set provided by an embodiment of the present application;

[0029] Figure 4 is a flowchart of a method for determining a second candidate search information set provided by an embodiment of the present application;

[0030] Figure 5 is a flowchart of an update method of a preset database provided by an embodiment of the present application;

[0031] Figure 6 is a flowchart of a method for determining a target search information group according to access frequency provided by an embodiment of the present application;

[0032] Figure 7 is a flowchart of a method for determining corresponding reply information according to a terminal's to-be-searched information provided by an embodiment of the present application;

[0033] Figure 8 is a display interface diagram of a search-reply information pair constructed by an embodiment of the present application;

[0034] Figure 9 is a display interface diagram of a recommended question provided by an embodiment of the present application;

[0035] Figure 10 is a search information clustering result diagram provided by the prior art;

[0036] Figure 11 is a search information clustering result diagram provided by an embodiment of the present application;

[0037] Figure 12 is a structural diagram of a reply information determination device provided by an embodiment of the present application;

[0038] Figure 13 Fig. 1 is a structural schematic diagram of a server provided by an embodiment of the present application. DETAILED DESCRIPTION

[0039] Artificial Intelligence (AI) is the use of digital computers or machines controlled by digital computers to simulate, extend and expand human intelligence, perceive the environment, acquire knowledge and use knowledge to obtain the best results. In other words, artificial intelligence is a comprehensive technology of computer science, which attempts to understand the essence of intelligence and produce a new intelligent machine that can react in a similar way to human intelligence. Artificial intelligence is to study the design principles and implementation methods of various intelligent machines, so that machines have the functions of perception, reasoning and decision-making.

[0040] Artificial intelligence technology is a comprehensive discipline, involving a wide range of fields, both hardware and software technologies. Artificial intelligence basic technologies generally include technologies such as sensors, special artificial intelligence chips, cloud computing, distributed storage, big data processing technology, operation / interaction system, mechatronics, etc. Artificial intelligence software technology mainly includes computer vision technology, speech processing technology, natural language processing technology, and machine learning / deep learning, automatic driving, intelligent transportation and several other major directions.

[0041] With the research and progress of artificial intelligence technology, artificial intelligence technology is researched and applied in many fields, such as common smart home, smart wearable devices, virtual assistants, smart speakers, smart marketing, unmanned vehicles, autonomous vehicles, drones, robots, intelligent medical care, intelligent customer service, Internet of Vehicles, autonomous driving, intelligent transportation, etc. It is believed that with the development of technology, artificial intelligence technology will be applied in more and more fields and play an increasingly important value.

[0042] Nature Language processing (NLP) is an important direction in the field of computer science and artificial intelligence. It studies various theories and methods that can realize effective communication between people and computers using natural language. Natural language processing is a science that integrates linguistics, computer science and mathematics. Therefore, the research in this field will involve natural language, i.e. the language used in daily life, so it is closely related to the study of linguistics. Natural language processing technology usually includes text processing, semantic understanding, machine translation, robot question and answer, knowledge graph and other technologies.

[0043] The scheme provided by the embodiments of the present application relates to the construction technology of question and answer pairs in the robot question and answer of artificial intelligence. The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative labor belong to the scope of protection of the present application.

[0044] It should be noted that the terms "first", "second" and the like in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and do not necessarily indicate a specific order or a chronological sequence. It should be understood that the data thus used can be interchanged under appropriate circumstances, so that the embodiments of the present application described herein can be implemented in an order other than that illustrated or described herein. In addition, the terms "include" and "have" and any variations thereof are intended to cover non-exclusive inclusion, for example, a process, method, system, product or server including a series of steps or units does not have to be limited to those steps or units clearly listed, but can include other steps or units not clearly listed or inherent to these processes, methods, products or devices.

[0045] Please refer to Figure 1 , Figure 1 is a schematic diagram of a reply information determination system provided by the embodiments of the present application, as Figure 1 shown, the reply information determination system can at least include a server 01 and a client 02.

[0046] Specifically, in the embodiments of the present application, the server 01 can include a standalone server, or a distributed server, or a server cluster composed of multiple servers, and can also be a cloud server providing cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, CDN (Content Delivery Network), and basic cloud computing services such as big data and artificial intelligence platforms. The server 01 can include a network communication unit, a processor, a memory and the like. Specifically, the server 01 can be used to construct a preset database, and determine the reply information corresponding to the target search information based on the preset database.

[0047] Specifically, in the embodiments of the present specification, the client 02 can include an entity device such as a smart phone, a desktop computer, a tablet computer, a notebook computer, a digital assistant, a smart wearable device, a smart speaker, a vehicle terminal, a smart television, etc., can include software running in the entity device, for example, a web page provided by a service provider to a user, and can be an application provided by the service provider to the user. Specifically, the client 02 can be used to input target search information.

[0048] A method for determining reply information is introduced below, Figure 2 is a flowchart of a method for determining reply information provided by an embodiment of the present application. The present specification provides method operation steps as described in the embodiments or flowcharts, but more or fewer operation steps can be included based on conventional or non-creative labor. The order of steps listed in the embodiments is only one of the many execution orders, and does not represent the only execution order. In actual system or server product execution, the method order shown in the embodiments or the drawings can be executed in sequence or in parallel (for example, in a parallel processor or multi-threaded processing environment). Specifically as shown in Figure 2 The method can include:

[0049] S201: In response to a trigger instruction for a target area, display target search information.

[0050] In the embodiments of the present specification, the target search information can be voice, text or picture information input by a development user of an AI intelligent product, and the search information is used to build a question and answer pair of the AI intelligent product.

[0051] The target search information can be designed by the development user, or can be obtained by collecting logs of user terminals of the AI intelligent product.

[0052] In some embodiments, the above method can further include:

[0053] Obtain target logs of a plurality of terminals; each target log is used to store search information reply records in each terminal;

[0054] Determine a set of un-replied search information based on the search information reply records in the plurality of target logs;

[0055] Determine any un-replied search information in the set of un-replied search information as the target search information.

[0056] In the embodiments of the present specification, the search information without reply information can be obtained through the logs of a plurality of product-side user terminals. Since the search information has not been replied, it is indicated that the corresponding reply information is not stored in the question and answer pair database of the intelligent product, and needs to be built by the development user.

[0057] S203: Obtain a preset database corresponding to the target search information; the preset database stores a corresponding relationship between search information and reply information.

[0058] In the embodiments of the present disclosure, the preset database is a database of intelligent products; the preset database can store the corresponding relationship between the search information and the reply information, wherein one piece of search information can correspond to one piece of reply information, and the same category of clustered search information also corresponds to the same piece of reply information, that is, the search information and the reply information can be in a one-to-one relationship or a many-to-one relationship.

[0059] In the embodiments of the present disclosure, each preset database can correspond to different fields, for example, can be divided into popular science field, music field, news field and the like. That is, a preset database corresponding to each field is constructed respectively.

[0060] In some embodiments, the obtaining of the preset database corresponding to the target search information comprises:

[0061] S2031: Determine a target field corresponding to the target search information;

[0062] S2033: Obtain a preset database corresponding to the target field.

[0063] In the embodiments of the present disclosure, by dividing the preset database by fields, the second candidate search information set can be quickly and accurately determined, so as to quickly determine the reply information of the target search information and improve the accuracy of the reply information.

[0064] In some embodiments, the method further comprises constructing the preset database by using the following method:

[0065] Obtain associated search information corresponding to the search information, the associated search information being information of the same intention as the search information;

[0066] Determine the reply information corresponding to the search information;

[0067] Take the reply information corresponding to the search information as the reply information of the associated search information;

[0068] Construct a search-reply information pair according to the search information, the associated search information and the reply information corresponding to the search information;

[0069] Construct the preset database based on a plurality of search-reply information pairs.

[0070] In the embodiments of this specification, the same response information can be set for search information and related search information with the same topic and intent. When constructing a new search-response information pair, similar search information to the target search information can be quickly recommended to the developer user, avoiding setting different response information for similar search information and improving the user's information query experience.

[0071] S205: Based on the deep learning algorithm, determine the first candidate search information set that matches the target search information from the above-mentioned preset database.

[0072] In the embodiments of this specification, the above-mentioned method of determining a first candidate search information set matching the target search information from the preset database based on a deep learning algorithm includes:

[0073] Based on deep learning algorithms, it is determined whether there is a first candidate search information in the aforementioned preset database that matches the aforementioned target search information;

[0074] If it exists, construct a first candidate search information set based on the first candidate search information that matches.

[0075] In this embodiment of the specification, if no first candidate search information matching the target search information exists in the preset database, the method further includes:

[0076] In response to a trigger command for a preset area, determine the response information corresponding to the target search information;

[0077] The preset database is updated based on the target search information and the corresponding response information.

[0078] In a specific embodiment, such as Figure 8 As shown, Figure 8 This is a diagram illustrating the display interface for constructing search-response information pairs in this embodiment. The search information can be a question, and the response information can be an answer. Developers can enter a question in the question box and an answer in the answer box, or the system can automatically match answers based on the entered question. Figure 9 As shown, Figure 9 This is a screenshot of the recommended questions interface. When a developer enters the question "How are the navigation lights of an aircraft distributed?", the system automatically recommends similar questions. The developer can select one or more of the recommended questions. For example, if questions 1 and 2 are selected, the corresponding answer "Aviation expert... to avoid collision between two aircraft" will be displayed in the answer box.

[0079] In the embodiments of the present specification, the preset area is used for the development user to input reply information, i.e., the standard answer corresponding to the target search information. When the target candidate information of the target search information does not exist in the preset database, the corresponding reply information needs to be input by the development user, so as to construct a search-reply information pair, and update the preset database based on the constructed search-reply information pair, thereby enriching the storage content in the preset database and improving the experience of the user using the intelligent product.

[0080] In the embodiments of the present specification, as shown in Figure 3 Based on the above-mentioned deep learning algorithm, the first candidate search information set matching the above-mentioned target search information is determined from the above-mentioned preset database, which includes:

[0081] S2051: Based on the above-mentioned deep learning algorithm, the first word vector feature of the above-mentioned target search information and the second word vector feature corresponding to each search information in the above-mentioned preset database are determined;

[0082] In the embodiments of the present specification, based on the above-mentioned deep learning algorithm, the first word vector feature of the above-mentioned target search information and the second word vector feature corresponding to each search information in the above-mentioned preset database are determined, which includes:

[0083] Based on the above-mentioned deep learning algorithm, a word vector feature determination model is constructed;

[0084] Based on the word vector feature determination model, the first word vector feature of the above-mentioned target search information and the second word vector feature corresponding to each search information in the above-mentioned preset database are determined.

[0085] In the embodiments of the present specification, the word vector feature determination model can be a Chinese pre-training language model roberta. The pre-training corpus data of this model is large, and the training is also quite sufficient, so the actual effect is very good, and bert-as-service is used to load the pre-trained roberta model, and then each search information query obtains the corresponding word vector feature by accessing the service. BERT (Bidirectional Encoder Representations from Transformer) service is an open source service, which allows users to use the BERT model in the form of calling services without paying attention to the implementation details of BERT.

[0086] S2052: Calculate the cosine similarity between the first word vector feature of the above-mentioned target search information and the second word vector feature corresponding to each search information;

[0087] In the embodiments of the present disclosure, the cosine similarity (cos value) between the first word vector feature and each second word vector feature can be calculated; the two word vector features with high cosine similarity correspond to search information with high similarity; and thus the candidate search information with high similarity to the target search information can be determined.

[0088] S2053: Determine the first candidate search information set based on the cosine similarity.

[0089] In the embodiments of the present disclosure, determining the first candidate search information set based on the cosine similarity includes:

[0090] S20531: Determine the search information corresponding to the second word vector feature with a cosine similarity greater than a similarity threshold value as the first candidate search information.

[0091] In the embodiments of the present disclosure, the similarity threshold value can be set according to actual needs, for example, it can be set to 0.85.

[0092] S20533: Determine the set composed of the first candidate search information as the first candidate search information set.

[0093] In the embodiments of the present disclosure, the word vector feature corresponding to the search information can be obtained according to the word vector feature determination model, and then the candidate search information with high similarity to the target search information is determined according to the cosine similarity between the word vector features, so as to obtain the first candidate search information set; the cosine similarity between the word vector features represents the theme and intent similarity between the word vector features.

[0094] S207: Determine the second candidate search information set matching the target search information from the first candidate search information set based on a string clustering algorithm.

[0095] In the embodiments of the present disclosure, the string clustering algorithm can be a K-means clustering (K means) algorithm; the K-means algorithm actually judges the proximity relationship of different samples by calculating the distance between them, and similar samples are put into the same category.

[0096] In the embodiments of the present disclosure, as shown in Figure 4 determining the second candidate search information set matching the target search information from the first candidate search information set based on a string clustering algorithm includes:

[0097] S2071: Obtain the character similarity between the first string corresponding to the target search information and the second string corresponding to each first candidate search information in the first candidate search information set.

[0098] In the embodiments of the present disclosure, by calculating the character similarity between the first string and the second string, the second candidate search information set that is closer to the topic and intent of the target search information can be screened from the first candidate search information set.

[0099] S2073: Determine the second candidate search information set based on the character similarity.

[0100] In the embodiments of the present disclosure, determining the second candidate search information set based on the character similarity includes:

[0101] S20731: Determine the first candidate search information corresponding to the second string with the character similarity greater than the preset threshold as the second candidate search information.

[0102] In the embodiments of the present disclosure, the preset threshold can be set according to actual needs, and the character similarity greater than the preset threshold indicates that the similarity of the two strings is high, that is, the topic and intent of the target search information are closer to the corresponding first candidate search information.

[0103] S20733: Determine the set composed of the second candidate search information as the second candidate search information set.

[0104] In the embodiments of the present disclosure, the second candidate search information in the second candidate search information set has a higher similarity to the target search information, which can be recommended to the development user for selection.

[0105] S209: Determine the target candidate information in response to the trigger instruction of at least one second candidate search information in the second candidate search information set.

[0106] In the embodiments of the present disclosure, the development user can select one or more target candidate information with high similarity to the topic and intent of the target search information, that is, high semantic similarity, based on the recommended second candidate search information set.

[0107] S2011: Determine the target reply information corresponding to the target candidate information based on the preset database, and take the target reply information as the reply information of the target search information.

[0108] In the embodiments of the present disclosure, after the target candidate information is determined, the corresponding target reply information is obtained and taken as the reply information of the target search information, thereby avoiding the development personnel to set new reply information for the target search information, thereby affecting the user experience. The present embodiment not only improves the user experience, but also saves the manpower, material resources and financial resources of the development end.

[0109] In the embodiments of the present disclosure, when the target candidate information is multiple, the method further comprises:

[0110] determining multiple target reply information of each target candidate information based on the preset database;

[0111] judging whether the multiple target reply information are same;

[0112] if the multiple target reply information are same, taking the target reply information as the reply information of the target search information.

[0113] In the embodiments of the present disclosure, if the multiple target candidate information of the same semantic information is developed by the user screening, and if the multiple target candidate information corresponds to multiple different reply information, the preset database needs to be updated.

[0114] In some embodiments, if the multiple target reply information are not same, the method further comprises:

[0115] in response to the trigger instruction, modifying the multiple target reply information into updated reply information;

[0116] taking the updated reply information as the reply information of the target search information.

[0117] In the embodiments of the present disclosure, if the multiple target reply information are not same, it indicates that multiple different answers of the same semantic question are stored in the preset database, and the developer needs to correct the multiple target reply information, and modify the multiple target reply information into updated reply information, and update the preset database according to the updated reply information, thereby improving the accuracy of the reply information in the preset database.

[0118] In the embodiments of the present disclosure, as shown in Figure 5 the method further comprises:

[0119] S501: obtaining target logs of multiple terminals; each target log is used to store search information reply records in each terminal;

[0120] In the embodiments of the present disclosure, the search information reply records can record the search information in the terminal and the corresponding reply information, and if there is no reply information, the search information reply records can also be recorded.

[0121] In some embodiments, the obtaining the target logs of the multiple terminals comprises:

[0122] sending a target log obtaining request to the multiple terminals based on identification information of the multiple terminals;

[0123] Receive multiple target logs sent by multiple terminals based on target log acquisition requests.

[0124] In the embodiments of this specification, since the amount of online log data is very large, not every terminal involves the preset database. Therefore, it is necessary to filter the terminals, select only the terminals that involve the preset database, and obtain their corresponding target logs.

[0125] S503: Based on the search information response records in multiple target logs, determine the set of unresponded search information;

[0126] In the embodiments of this specification, in order to improve the user experience on the product side, search information that has not received a reply can be collected, and corresponding reply information can be set for it; and the preset database can be updated. In this way, when the user enters search information again, a reply information can be obtained, thereby improving the user experience.

[0127] S505: Based on the above deep learning algorithm and the above string clustering algorithm, the above unanswered search information set is clustered to obtain multiple search information groups;

[0128] In the embodiments of this specification, the search information in each search information group has the same theme and intent. For example, regarding the question of why penguins cannot fly, the corresponding search information group includes: "Why can't penguins fly?", "Why can't penguins fly?", "Why can't penguins fly even though they have wings?", "Why can't penguins fly even though they are birds?", "Jingle bells, why can't penguins fly?", etc.

[0129] In the embodiments of this specification, based on the aforementioned deep learning algorithm and string clustering algorithm, the aforementioned unanswered search information set is clustered to obtain multiple search information groups, including:

[0130] S5051: Based on the above deep learning algorithm, determine the word vector feature set of each unanswered search message in the unanswered search information set; and use the word vector feature set as the current word vector feature set;

[0131] S5053: Use any one of the word vector features in the current word vector feature set as the initial feature;

[0132] S5055: Calculate the cosine similarity between the initial feature and each word vector feature in the current word vector feature set, excluding the initial feature;

[0133] S5057: Based on the above string clustering algorithm, word vector features with cosine similarity less than the preset similarity threshold are removed from the current word vector feature set to obtain clustered word vector feature groups with cosine similarity greater than or equal to the preset similarity threshold.

[0134] S5059: based on the removed word vector features, a remaining word vector feature set is formed; and the remaining word vector feature set is re-taken as the current word vector feature set;

[0135] S50511: if the number of features in the current word vector feature set is greater than the target number, the above step of taking any word vector feature in the current word vector feature set as an initial feature to form a remaining word vector feature set based on the removed word vector features is repeated; and the remaining word vector feature set is re-taken as the current word vector feature set;

[0136] S50513: based on the plurality of clustered word vector feature groups, a plurality of search information groups is obtained.

[0137] In the embodiments of the present disclosure, each word vector feature in the clustered word vector feature group corresponds to a search information group.

[0138] In the embodiments of the present disclosure, after the word vector features with a cosine similarity less than a preset similarity threshold are removed from the above current word vector feature set, the un-replied search information corresponding to the word vector features with a cosine similarity greater than or equal to the preset similarity threshold are automatically clustered into a category; and the same method is used to cluster the remaining word vector features, thereby obtaining a plurality of search information groups.

[0139] S507: determining a reply information corresponding to each search information group.

[0140] In the embodiments of the present disclosure, each search information group is search information with the same theme and intent, and thus each search information group corresponds to a reply information, thereby avoiding setting different reply information for search information with the same theme and intent, improving user experience, and reducing the cost of manpower, material resources and financial resources.

[0141] In some embodiments, as shown in FIG. 13, before the above step of determining a reply information corresponding to each search information group, the above method further includes: Figure 6

[0142] S5061: determining an access frequency of each search information group;

[0143] In the embodiments of the present disclosure, the access frequency can be represented by page views (pv).

[0144] In some embodiments, the above step of determining an access frequency of each search information group includes:

[0145] obtaining the number of search information in each search information group and the access frequency of each search information in each search information group;

[0146] ​According to the number of search information in each search information group and the access frequency of each search information in each search information group, the access frequency of each search information group is calculated.

[0147] In the embodiments of the present disclosure, the product of the number of search information in each search information group and the access frequency of each search information in each search information group can be taken as the access frequency of each search information group; for example, there are two search information in a search information group, one with a frequency of 1 and the other with a frequency of 9, and the access frequency of the whole group, that is, the page view (pv) is 10. The embodiments realize determining the access frequency of each search information group according to the number of search information in each search information group and the access frequency of each search information in the group, thereby determining the importance of each search information group.

[0148] S5063: determining the target search information group based on the access frequency.

[0149] Correspondingly, the determining of the reply information corresponding to each search information group includes:

[0150] The reply information corresponding to the target search information group is determined.

[0151] In the embodiments of the present disclosure, the high-frequency and important target search information can be determined according to the access frequency of search information in the search information group, and the reply information of this part of search information is determined preferentially; the high access frequency indicates that the target search information is relatively important, and the corresponding reply information needs to be determined preferentially, thereby the user loss can be avoided to the maximum extent.

[0152] In some embodiments, after the access frequency of each search information group is determined, the method further includes:

[0153] Based on the access frequency, the plurality of search information groups are sorted;

[0154] The determining of the reply information corresponding to each search information group includes:

[0155] The reply information corresponding to each search information group is determined in turn according to the order from large to small of the access frequency.

[0156] In the embodiments of the present disclosure, the search information groups can be sorted according to the access frequency, and the reply information is determined in turn according to the sorting result, thereby ensuring that the important and high-frequency search information is processed preferentially.

[0157] S509: updating the preset database based on the search information in each search information group and the reply information corresponding to each search information group.

[0158] In the embodiments of the present application, the newly added search-reply information pairs can be constructed according to the search information in each search information group and the reply information corresponding to each search information group, and the newly added search-reply information pairs are stored in the preset database, so as to update the preset database and improve the information search experience of the user.

[0159] In some embodiments, as shown in Figure 7 The method further includes:

[0160] S701: receiving the to-be-searched information sent by the terminal;

[0161] In the embodiments of the present application, the terminal is a user terminal of an intelligent product, and the to-be-searched information is information input by a user of the product terminal, which can be text information converted from voice information.

[0162] S703: determining a target search information set matched with the to-be-searched information from the preset database based on a deep learning algorithm.

[0163] S705: determining a candidate search information set matched with the to-be-searched information from the target search information set based on a string clustering algorithm.

[0164] In the embodiments of the present application, the string clustering algorithm can be a K-means algorithm.

[0165] S707: determining candidate reply information corresponding to the candidate search information based on the preset database, and taking the candidate reply information as the reply information of the to-be-searched information.

[0166] S709: sending the candidate reply information to the terminal.

[0167] In the embodiments of the present application, the user terminal of the intelligent product can be provided with an updated preset database. If the search information input by the user is stored in the preset database, the corresponding reply information can be directly found. If the search information input by the user is not stored in the preset database, the candidate search information matched with the to-be-searched information input by the user needs to be determined based on the deep learning algorithm and the string clustering algorithm, and the candidate reply information is determined. In this way, the problem that the user cannot be replied to when the to-be-searched information is not stored in the preset database can be avoided, the probability of being unable to reply to the information is reduced, and the user experience is improved.

[0168] In the embodiments of the present application, the search information in the same batch of search information sets is clustered by using the K-means algorithm of the prior art and the method of the present application respectively, and the clustering result schematic diagram is obtained. Figures 10-11 Figure 10 ​As shown, many are not a class of inductive into a category, for example, the semantic gap between the search information of class ID 2 is large, and the clustering effect is poor. Figure 11 As shown, the accuracy of clustering is greatly improved, and in addition, the high-frequency problems are given very high priority by using the ranking results of class PV.

[0169] From the technical solutions provided by the above embodiments of the present specification, it can be seen that in the process of constructing the question and answer pair, when the target search information is developed by the user, the deep learning algorithm and the string clustering algorithm are used in sequence to determine a set of to-be-recommended search information (i.e., a second candidate search information set) that is similar to the theme and intent of the target search information. The user can select a target candidate information that is the same as the theme and intent of the target search information based on the second candidate search information set. The similarity between the to-be-recommended search information set and the target search information is improved, thereby facilitating the user to quickly determine the target candidate information that is the same as the theme and intent of the target search information. The present application takes the target candidate information corresponding to the target reply information as the reply information of the target search information, effectively avoids setting different reply information for similar search information, improves the accuracy of the reply information, and improves the information query experience of the user.

[0170] The present application also provides a reply information determination device, as shown in Figure 12 The device comprises:

[0171] The target search information display module 1210 is configured to display the target search information in response to a trigger instruction for the target area.

[0172] The preset database acquisition module 1220 is configured to acquire a preset database corresponding to the target search information. The preset database stores the correspondence between the search information and the reply information.

[0173] The first candidate search information set determination module 1230 is configured to determine a first candidate search information set matching the target search information from the preset database based on a deep learning algorithm.

[0174] The second candidate search information set determination module 1240 is configured to determine a second candidate search information set matching the target search information from the first candidate search information set based on a string clustering algorithm.

[0175] The target candidate information determination module 1250 is configured to determine a target candidate information in response to a trigger instruction for at least one second candidate search information in the second candidate search information set.

[0176] The target reply information determination module 1260 is configured to determine target reply information corresponding to the target candidate information based on the preset database, and take the target reply information as the reply information of the target search information.

[0177] In some embodiments, the first candidate search information set determination module can include:

[0178] The word vector feature determination unit is configured to determine a first word vector feature of the target search information and a second word vector feature corresponding to each search information in the preset database based on the deep learning algorithm.

[0179] The cosine similarity calculation unit is configured to calculate a cosine similarity between the first word vector feature of the target search information and the second word vector feature corresponding to each search information.

[0180] The first candidate search information set is determined based on the cosine similarity.

[0181] In some embodiments, the second candidate search information set determination module can include:

[0182] The character similarity acquisition unit is configured to acquire a character similarity between a first character string corresponding to the target search information and a second character string corresponding to each first candidate search information in the first candidate search information set.

[0183] The second candidate search information set determination unit is configured to determine a second candidate search information set based on the character similarity.

[0184] In some embodiments, the apparatus can further include:

[0185] The target log acquisition module is configured to acquire target logs of a plurality of terminals, and each target log is configured to store search information reply records in each terminal.

[0186] The unreplied search information set determination module is configured to determine an unreplied search information set based on the search information reply records in the plurality of target logs.

[0187] The search information group determination module is configured to cluster the unreplied search information set based on the deep learning algorithm and the character string clustering algorithm to obtain a plurality of search information groups.

[0188] The reply information determination module is configured to determine reply information corresponding to each search information group.

[0189] The update module is configured to update the preset database based on the search information in each search information group and the reply information corresponding to each search information group.

[0190] In some embodiments, the search information group determination module can comprise:

[0191] a word vector feature set determination unit configured to determine, based on the deep learning algorithm, a word vector feature set of each un-replied search information in the un-replied search information set, and take the word vector feature set as a current word vector feature set;

[0192] an initial feature determination unit configured to take any word vector feature in the current word vector feature set as an initial feature;

[0193] a cosine similarity calculation unit configured to calculate a cosine similarity between the initial feature and each word vector feature in the current word vector feature set except the initial feature;

[0194] a clustered word vector feature group determination unit configured to, based on the string clustering algorithm, eliminate word vector features with a cosine similarity less than a preset similarity threshold from the current word vector feature set to obtain a clustered word vector feature group;

[0195] a remaining word vector feature set determination unit configured to, based on the eliminated word vector features, form a remaining word vector feature set, and take the remaining word vector feature set as the current word vector feature set again;

[0196] a repeating unit configured to, if the number of features in the current word vector feature set is greater than a target number, repeat the steps of taking any word vector feature in the current word vector feature set as an initial feature, based on the eliminated word vector features, forming a remaining word vector feature set, and taking the remaining word vector feature set as the current word vector feature set again;

[0197] a plurality of search information group determination units configured to, based on a plurality of clustered word vector feature groups, obtain a plurality of search information groups.

[0198] In some embodiments, the device can further comprise:

[0199] a visit frequency determination module configured to determine a visit frequency of each search information group;

[0200] a target search information group determination module configured to determine a target search information group based on the visit frequency;

[0201] Correspondingly, the reply information determination module comprises:

[0202] a reply information determination unit configured to determine reply information corresponding to the target search information group.

[0203] In some embodiments, the device can further comprise:

[0204] The sorting module is configured to sort the plurality of search information groups based on the access frequencies.

[0205] In some embodiments, the reply information determination unit can include:

[0206] The reply information determination sub-unit is configured to determine, in order, the reply information corresponding to each search information group according to the order from high to low of the access frequencies.

[0207] In some embodiments, the access frequency determination module can include:

[0208] The access frequency acquisition unit is configured to acquire the number of search information in each search information group and the access frequency of each search information in each search information group.

[0209] The access frequency calculation unit is configured to calculate the access frequency of each search information group according to the number of search information in each search information group and the access frequency of each search information in each search information group.

[0210] In some embodiments, the apparatus can further include:

[0211] The to-be-searched information receiving module is configured to receive the to-be-searched information sent by the terminal.

[0212] The candidate search information determination module is configured to determine, from the preset database, candidate search information based on the deep learning algorithm and the string clustering algorithm.

[0213] The target search information set determination module is configured to determine, from the preset database, a target search information set matching the to-be-searched information based on the deep learning algorithm.

[0214] The candidate reply information determination module is configured to determine, from the target search information set, a candidate search information set matching the to-be-searched information based on the string clustering algorithm.

[0215] The candidate reply information sending module is configured to send the candidate reply information to the terminal.

[0216] The apparatus in the apparatus embodiment and the method embodiment are based on the same inventive concept.

[0217] The embodiments of the present application provide a reply information determination device, which includes a processor and a memory, the memory storing at least one instruction or at least one program, the at least one instruction or at least one program being loaded and executed by the processor to implement the reply information determination method provided in the above method embodiments.

[0218] The embodiment of the present application further provides a computer storage medium, which can be arranged in a terminal to store at least one instruction or at least one program related to a reply information determination method in the method embodiment, and the at least one instruction or at least one program is loaded and executed by the processor to implement the reply information determination method provided in the above method embodiment.

[0219] The embodiment of the present application further provides a computer program product or computer program, which comprises computer instructions stored in a computer readable storage medium. The processor of the computer device reads the computer instructions from the computer readable storage medium, and the processor executes the computer instructions, so that the computer device executes the method provided in the various optional implementation manners.

[0220] Optionally, in the embodiment of the present application, the storage medium can be located in at least one network server of a plurality of network servers of a computer network. Optionally, in the embodiment, the storage medium can include but is not limited to a U disk, a read-only memory (ROM), a random access memory (RAM), a mobile hard disk, a magnetic disk or an optical disk, and various media capable of storing program codes.

[0221] The memory in the embodiment of the present application can be used to store software programs and modules, and the processor executes various function applications and data processing by running the software programs and modules stored in the memory. The memory can mainly include a program storage area and a data storage area, wherein the program storage area can store an operating system, application programs required by functions, etc.; and the data storage area can store data created according to the use of the device, etc. In addition, the memory can include a high-speed random access memory, and can also include a non-volatile memory, for example, at least one magnetic disk storage device, a flash memory device or other volatile solid-state memory device. Accordingly, the memory can also include a memory controller to provide the processor with access to the memory.

[0222] The reply information determination method embodiment provided by the embodiment of the present application can be executed in a mobile terminal, a computer terminal, a server or a similar computing device. Taking the case of running on a server as an example, Figure 13 is a hardware structure block diagram of a server of a reply information determination method provided by the embodiment of the present application. As shown in Figure 13As shown, the server 1300 can vary greatly in configuration and performance, and can include one or more Central Processing Units (CPU) 1310 (processor(s) 1310 can include, without limitation, a microprocessor, an Application Specific Integrated Circuit (ASIC), a Field-Programmable Gate Array (FPGA), or the like), memory 1330 for storing data, one or more storage media 1320 (such as one or more mass storage devices) for storing applications 1323 or data 1322. The memory 1330 and the storage media 1320 can be of any type of volatile or non-volatile storage, including, without limitation, magnetic storage, semiconductor storage, or the like. The storage media 1320 can include one or more storage devices, including, without limitation, a hard disk drive, a floppy disk drive, a CD-ROM, a DVD-ROM, a Blu-ray disk, a tape drive, a flash memory device, or the like. The storage media 1320 can be external to the server 1300, or internal to the server 1300. The storage media 1320 can be connected to the server 1300 via a wired or wireless connection. The server 1300 can also include one or more Input / Output (I / O) interfaces 1340, one or more network interfaces 1350, and / or one or more operating systems 1321, such as Windows Server™, Mac OS X™, Unix™, Linux™, FreeBSD™, or the like.

[0223] The I / O interface 1340 can be used to receive or send data via a network. Examples of the network can include a wireless network provided by a communication provider of the server 1300. In one example, the I / O interface 1340 includes a Network Interface Controller (NIC) that can connect to other network devices through a base station to communicate with the Internet. In one example, the I / O interface 1340 can be a Radio Frequency (RF) module that is used to communicate with the Internet through a wireless manner.

[0224] Those of ordinary skill in the art can understand, Figure 13 The structure shown is merely illustrative, and does not limit the structure of the electronic device described above. For example, the server 1300 can include more or less components than shown, or have a different configuration of components than shown. Figure 13 The structure shown is merely illustrative, and does not limit the structure of the electronic device described above. For example, the server 1300 can include more or less components than shown, or have a different configuration of components than shown. Figure 13

[0225] ​According to the embodiments of the method, the device, the equipment or the storage medium for determining reply information provided in the application, when developing target search information, the deep learning algorithm and the string clustering algorithm are used to determine the set of to-be-recommended search information (i.e., the second candidate search information set) that is similar to the theme and the intention of the target search information, so that the development user can select the target candidate information that is the same as the theme and the intention of the target search information based on the second candidate search information set. The similarity between the set of to-be-recommended search information and the target search information is improved, so that the development user can quickly determine the target candidate information that is the same as the theme and the intention of the target search information. The target reply information corresponding to the target candidate information is used as the reply information of the target search information, the different reply information for similar search information is effectively avoided, the accuracy of the reply information is improved, and the information query experience of the user is improved.

[0226] It should be noted that the above-mentioned embodiments of the application are only for description, and do not represent the advantages and disadvantages of the embodiments. The above describes a specific embodiment of the present application. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recorded in the claims can be executed in an order different from that in the embodiments and still achieve the desired results. In addition, the processes depicted in the drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multi-task processing and parallel processing are possible or can be advantageous.

[0227] Each embodiment in the specification is described in a progressive manner, and the same or similar parts between each embodiment can be referred to each other. Each embodiment focuses on the difference from other embodiments. In particular, for the device, equipment and storage medium embodiments, since they are basically similar to the method embodiments, the description is relatively simple, and the relevant parts are referred to the part of the method embodiment.

[0228] Those of ordinary skill in the art can understand that all or part of the steps of the above-mentioned embodiments can be completed by hardware, or by a program instructing relevant hardware to complete, and the program can be stored in a computer storage medium. The storage medium mentioned above can be a read-only memory, a magnetic disk or an optical disk.

[0229] The above is only the preferred embodiment of the application, and does not limit the application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the application shall be included in the protection scope of the application.

Claims

1. A method for determining response information, characterized in that, The method includes: In response to a trigger command for the target area, display target search information; Obtain a preset database corresponding to the target search information; the preset database stores the correspondence between search information and response information. Based on deep learning algorithms, a first set of candidate search information that matches the target search information is determined from the preset database; Based on a string clustering algorithm, a second set of candidate search information that matches the target search information is determined from the first set of candidate search information. In response to a triggering instruction on at least one second candidate search information in the second candidate search information set, the target candidate information is determined; Based on the preset database, the target response information corresponding to the target candidate information is determined; and the target response information is used as the response information for the target search information.

2. The method according to claim 1, characterized in that, The step of determining a first candidate search information set matching the target search information from the preset database based on a deep learning algorithm includes: Based on the deep learning algorithm, the first word vector features of the target search information and the second word vector features corresponding to each search information in the preset database are determined. Calculate the cosine similarity between the first word vector feature of the target search information and the second word vector feature corresponding to each search information; Based on the cosine similarity, the first candidate search information set is determined.

3. The method according to claim 1, characterized in that, The step of determining a second set of candidate search information that matches the target search information from the first set of candidate search information based on the string clustering algorithm includes: Obtain the character similarity between the first string corresponding to the target search information and the second string corresponding to each first candidate search information in the first candidate search information set; Based on the character similarity, a second candidate search information set is determined.

4. The method according to claim 1, characterized in that, The method further includes: Acquire target logs from multiple terminals; each target log is used to store search information response records from each terminal. Based on the search information response records in multiple target logs, determine the set of search information that has not been responded to; Based on the deep learning algorithm and the string clustering algorithm, the unanswered search information set is clustered to obtain multiple search information groups; Determine the response information corresponding to each search information group; The preset database is updated based on the search information in each search information group and the corresponding response information for each search information group.

5. The method according to claim 4, characterized in that, The method, based on the deep learning algorithm and the string clustering algorithm, clusters the unanswered search information set to obtain multiple search information groups, including: Based on the deep learning algorithm, the word vector feature set of each unanswered search message in the unanswered search message set is determined; and the word vector feature set is used as the current word vector feature set. Use any one of the word vector features in the current word vector feature set as the initial feature; Calculate the cosine similarity between the initial feature and each word vector feature in the current word vector feature set, excluding the initial feature; Based on the string clustering algorithm, word vector features with cosine similarity less than a preset similarity threshold are removed from the current word vector feature set to obtain clustered word vector feature groups; Based on the removed word vector features, a remaining word vector feature set is formed; and the remaining word vector feature set is then used as the current word vector feature set. If the number of features in the current word vector feature set is greater than the target number, repeat the steps: take any word vector feature in the current word vector feature set as the initial feature and combine it with the word vector features based on elimination to form the remaining word vector feature set; and use the remaining word vector feature set as the current word vector feature set again. Multiple search information groups are obtained based on multiple clustered word vector feature groups.

6. The method according to claim 4, characterized in that, Before determining the response information corresponding to each search information group, the method further includes: Determine the access frequency for each search information group; Based on the access frequency, the target search information group is determined; Accordingly, determining the response information corresponding to each search information group includes: Determine the response information corresponding to the target search information group.

7. The method according to claim 6, characterized in that, After determining the access frequency of each search information group, the method further includes: The multiple search information groups are sorted based on the access frequency; Determining the response information corresponding to each search information group includes: The response information for each search information group is determined sequentially according to the order of access frequency from highest to lowest.

8. The method according to claim 6, characterized in that, Determining the access frequency of each search information group includes: Obtain the number of search items in each search information group and the access frequency of each search item in each search information group; The access frequency of each search information group is calculated based on the number of search information in each search information group and the access frequency of each search information in each search information group.

9. The method according to any one of claims 1-8, characterized in that, The method further includes: The search information sent by the receiving terminal; Based on the deep learning algorithm, a target search information set that matches the information to be searched is determined from the preset database; Based on the string clustering algorithm, a set of candidate search information that matches the information to be searched is determined from the target search information set; Based on the preset database, candidate response information corresponding to the candidate search information is determined; and the candidate response information is used as the response information for the information to be searched. The candidate response information is sent to the terminal.

10. A response information determining device, characterized in that, The device includes: The target search information display module is used to display target search information in response to a trigger command on the target area; A preset database acquisition module is used to acquire a preset database corresponding to the target search information; the preset database stores the correspondence between search information and response information. The first candidate search information set determination module is used to determine a first candidate search information set that matches the target search information from the preset database based on a deep learning algorithm. The second candidate search information set determination module is used to determine a second candidate search information set that matches the target search information from the first candidate search information set based on a string clustering algorithm. The target candidate information determination module is used to determine target candidate information in response to a triggering instruction for at least one second candidate search information in the second candidate search information set; The target response information determination module is used to determine the target response information corresponding to the target candidate information based on the preset database; and to use the target response information as the response information of the target search information.

Images