CN116414941A - Information query method, device, equipment and storage medium - Google Patents

Abstract

The invention discloses an information query method, device, equipment and storage medium, belonging to the technical field of the Internet. The method includes: when receiving a query instruction, determining query data according to the query instruction; determining query data according to the query data text and query vector; perform keyword retrieval according to the query text, and perform vector retrieval according to the query vector; combine keyword retrieval results and vector retrieval results to determine information query results. This scheme uses keyword retrieval and vector retrieval to obtain the retrieval results respectively, and then synthesizes the retrieval results to obtain the information query results, which can generalize the query results and achieve the purpose of generalized query, so as to make full use of the data and more efficiently query the information Carry out generalization, improve the effect of information query, and better meet the needs of users.

Description

Information query method, device, equipment and storage medium

technical field

The present invention relates to the technical field of the Internet, in particular to an information query method, device, equipment and storage medium.

Background technique

In search, recommendation and advertising systems, there are often scenarios where specific content needs to be displayed when certain texts need to be precisely hit. In this scenario, accuracy is usually pursued, and it usually takes effect only when it is precisely hit. However, how to generalize the hit text (generally query in search, hereinafter collectively referred to as query) is a difficult problem.

The existing technology is basically divided into different query types, and different methods are used for different types of queries, and the offline mining method that is often used in the extension, the online effect is still through the full hit or full hit rule. For example, use strategies such as translation models or co-clicks to mine query candidate sets, then use algorithms such as semantic matching to control the relevance, and finally put the mined queries online. This technical solution has problems such as poor timeliness, poor scalability, and limited generalization ability. Because the query mapping relationship needs to be processed offline, it is still impossible to match data that has not been seen, and different types are often used. The mining means and methods cannot be applied to the full amount of data.

The above content is only used to assist in understanding the technical solution of the present invention, and does not mean that the above content is admitted as prior art.

Contents of the invention

The main purpose of the present invention is to propose an information query method, device, equipment and storage medium, aiming to solve the technical problem of how to make full use of data, more efficiently generalize the queried information, and improve the effect of information query.

In order to achieve the above object, the present invention provides an information query method, the information query method comprising:

When receiving a query instruction, determine query data according to the query instruction;

determining a query text and a query vector according to the query data;

performing keyword retrieval according to the query text, and performing vector retrieval according to the query vector;

Combining keyword retrieval results and vector retrieval results to determine information query results.

Optionally, performing keyword retrieval according to the query text, and performing vector retrieval according to the query vector include:

Keyword retrieval is performed according to the query text and a preset keyword index, and vector retrieval is performed according to the query vector and a preset vector index.

Optionally, before performing the keyword retrieval according to the query text and the preset keyword index, and performing the vector retrieval according to the query vector and the preset vector index, the method further includes:

Obtain sample data of various business types;

generating a candidate set to be matched according to the sample data;

Generate a preset keyword index and a preset vector index according to the candidate set to be matched.

Optionally, the generating a preset keyword index and a preset vector index according to the candidate set to be matched includes:

Generate the offline vector corresponding to the candidate set to be matched by preset deep learning representation model;

A preset keyword index is generated according to the candidate set to be matched, and a preset vector index is generated according to the offline vector.

Optionally, before generating the offline vector corresponding to the candidate set to be matched through the preset deep learning representation model, it also includes:

Obtain the first training data;

The initial deep learning model is trained according to the first training data to obtain a preset deep learning representation model.

Optionally, performing keyword retrieval according to the query text and a preset keyword index, and performing vector retrieval according to the query vector and a preset vector index include:

Configure the search engine according to the preset keyword index and preset vector index;

inputting said query text and said query vector into said search engine;

The search engine calls a preset keyword index to perform keyword search on the query text, and the search engine calls a preset vector index to perform vector search on the query vector.

Optionally, the determining the query text and the query vector according to the query data includes:

Respectively performing demand identification processing and vectorization processing according to the query data;

The query text is determined according to the requirement identification processing result, and the query vector is determined according to the vectorization processing result.

Optionally, performing demand identification processing and vectorization processing respectively according to the query data includes:

performing demand identification processing on the query data through a demand identification technology;

The query data is vectorized through a preset deep learning representation model.

Optionally, the combination of keyword retrieval results and vector retrieval results to determine information query results includes:

determining first retrieval information corresponding to the query data according to the keyword retrieval results;

determining second retrieval information corresponding to the query data according to the vector retrieval result;

generating target retrieval information according to the first retrieval information and the second retrieval information;

An information query result is generated according to the target retrieval information.

Optionally, the generating target retrieval information according to the first retrieval information and the second retrieval information includes:

Combining the first search information and the second search information to determine a set of the first search information and the second search information;

Target retrieval information is generated according to a set of the first retrieval information and the second retrieval information.

Optionally, after determining the information query result by combining the keyword search result and the vector search result, it further includes:

Performing a semantic correlation analysis on the information query results through a preset deep learning interactivity model;

Detect semantic consistency based on semantic correlation analysis results;

The target information query result is obtained according to the detection result, and the target information query result is displayed.

Optionally, before performing semantic correlation analysis on the information query results through the preset deep learning interactivity model, it also includes:

Obtain second training data;

The initial deep learning model is trained according to the second training data to obtain a preset deep learning interactive model.

In addition, in order to achieve the above purpose, the present invention also proposes an information query device, which includes:

A query data module, configured to determine query data according to the query command when the query command is received;

A text vector module, configured to determine query text and query vectors according to the query data;

An information retrieval module, configured to perform keyword retrieval according to the query text, and perform vector retrieval according to the query vector;

The query result module is used to determine the information query result in combination with the keyword retrieval result and the vector retrieval result.

Optionally, the information retrieval module is further configured to perform keyword retrieval according to the query text and a preset keyword index, and perform vector retrieval according to the query vector and a preset vector index.

Optionally, the information query device also includes:

The index generation module is used to obtain sample data of various business types; generate a candidate set to be matched according to the sample data; generate a preset keyword index and a preset vector index according to the candidate set to be matched.

Optionally, the index generating module is further configured to generate an offline vector corresponding to the candidate set to be matched through a preset deep learning representation model; generate a preset keyword index according to the candidate set to be matched, and generate a preset keyword index according to the The above offline vectors generate preset vector indexes.

Optionally, the information query device also includes:

The model training module is used to acquire first training data; train the initial deep learning model according to the first training data to obtain a preset deep learning representation model.

Optionally, the information retrieval module is also configured to configure a retrieval engine according to a preset keyword index and a preset vector index; input the query text and the query vector into the retrieval engine; call The preset keyword index performs keyword search on the query text, and the search engine invokes a preset vector index to perform vector search on the query vector.

In addition, in order to achieve the above object, the present invention also proposes an information query device, the information query device includes: a memory, a processor, and an information query program stored in the memory and operable on the processor. When the above information query program is executed by the processor, the above information query method is realized.

In addition, in order to achieve the above purpose, the present invention also proposes a storage medium, on which an information query program is stored, and when the information query program is executed by a processor, the above information query method is realized.

In the information query method proposed by the present invention, when a query instruction is received, the query data is determined according to the query instruction; the query text and the query vector are determined according to the query data; keyword retrieval is performed according to the query text, and according to the The above query vector is used for vector retrieval; and the information query result is determined by combining the keyword retrieval result and the vector retrieval result. This scheme uses keyword retrieval and vector retrieval to obtain the retrieval results respectively, and then synthesizes the retrieval results to obtain the information query results, which can generalize the query results and achieve the purpose of generalized query, so as to make full use of the data and more efficiently query the information Carry out generalization, improve the effect of information query, and better meet the needs of users.

Description of drawings

Fig. 1 is a schematic structural diagram of an information query device of a hardware operating environment involved in the solution of an embodiment of the present invention;

Fig. 2 is a schematic flow chart of the first embodiment of the information query method of the present invention;

Fig. 3 is the overall flowchart of the generalized query of an embodiment of the information query method of the present invention;

Fig. 4 is a schematic flow chart of the second embodiment of the information query method of the present invention;

Fig. 5 is a schematic flow chart of the third embodiment of the information query method of the present invention;

Fig. 6 is a schematic diagram of functional modules of the first embodiment of the information query device of the present invention.

The realization of the purpose of the present invention, functional characteristics and advantages will be further described in conjunction with the embodiments and with reference to the accompanying drawings.

Detailed ways

It should be understood that the specific embodiments described here are only used to explain the present invention, not to limit the present invention.

Referring to FIG. 1 , FIG. 1 is a schematic structural diagram of an information query device in a hardware operating environment involved in the solution of an embodiment of the present invention.

As shown in FIG. 1 , the information query device may include: a processor 1001 , such as a central processing unit (Central Processing Unit, CPU), a communication bus 1002 , a user interface 1003 , a network interface 1004 , and a memory 1005 . Wherein, the communication bus 1002 is used to realize connection and communication between these components. The user interface 1003 may include a display screen (Display) and an input unit such as a button, and the optional user interface 1003 may also include a standard wired interface and a wireless interface. Optionally, the network interface 1004 may include a standard wired interface and a wireless interface (such as a Wi-Fi interface). The memory 1005 may be a high-speed random access memory (Random Access Memory, RAM), or a stable memory (non-volatile memory), such as a disk memory. Optionally, the memory 1005 may also be a storage device independent of the aforementioned processor 1001 .

Those skilled in the art can understand that the device structure shown in FIG. 1 does not constitute a limitation to the information query device, and may include more or less components than those shown in the figure, or combine some components, or arrange different components.

As shown in FIG. 1 , the memory 1005 as a storage medium may include an operating system, a network communication module, a user interface module, and an information query program.

In the information query device shown in Figure 1, the network interface 1004 is mainly used to connect to the external network and perform data communication with other network devices; the user interface 1003 is mainly used to connect to user equipment and perform data communication with the user equipment; the present invention The device invokes the information query program stored in the memory 1005 through the processor 1001, and executes the information query method provided by the embodiment of the present invention.

Based on the above hardware structure, an embodiment of the information query method of the present invention is proposed.

Referring to FIG. 2 , FIG. 2 is a schematic flowchart of the first embodiment of the information query method of the present invention.

In the first embodiment, the information query method includes:

Step S10, when a query instruction is received, determine query data according to the query instruction.

It should be noted that the execution subject of this embodiment can be an information query device, and the information query device can be a computer device with a data processing function, or other devices that can achieve the same or similar functions, and this embodiment does not make any In this embodiment, an information query device is used as an example for illustration.

It should be noted that the information query in this embodiment may include but not limited to text query, and may also include other types of information query, which is not limited in this embodiment. In this embodiment, text query is used as an example for illustration .

It should be noted that the query data in this embodiment refers to the query condition data input by the user, and the query data may include but not limited to text data, and may also include other types of data, which is not limited in this embodiment. In this embodiment, text data is taken as an example for description.

It should be understood that when a user needs to query information, he can input query data in the query interface, and then click a query button to send a query instruction to the computer device. When the computing device receives the query instruction, it can determine the corresponding query data according to the query instruction, and then determine the information query result according to the query data.

In a specific implementation, for example, assuming that the user wants to inquire about the weather tomorrow, the query data is "how is the weather tomorrow" input by the user, and the finally obtained information query result may be "tomorrow will be sunny to cloudy".

Step S20, determining a query text and a query vector according to the query data.

It should be noted that, different from the prior art, this solution determines the two retrieval results in two different ways according to the query text and the query vector, and then combines the two retrieval results to obtain the information query result, so that the information The query results are generalized to obtain more extensive information query results to better meet the needs of users.

It should be understood that requirement recognition processing and vectorization processing may be performed on the query data respectively, so as to obtain query text and query vectors.

It is understandable that when retrieval is required, the preset deep learning representation model can be used to vectorize the query data to obtain a query vector, and at the same time, the query demand recognition technology can be used to perform some standardized processing on the query data to obtain the query text.

Step S30, perform keyword retrieval according to the query text, and perform vector retrieval according to the query vector.

It should be noted that the preset keyword index and preset vector index can be generated in advance according to the candidate set to be matched, and after the query text and query vector are determined in the above-mentioned way, keyword retrieval can be performed according to the query text and preset keyword index , to select the data related to the query text from the candidate set to be matched to obtain the keyword retrieval result. At the same time, vector retrieval can also be performed according to the query vector and the preset vector index, so as to select data related to the query vector from the candidate set to be matched to obtain the vector retrieval result.

Step S40, combining the keyword retrieval result and the vector retrieval result to determine the information query result.

It should be understood that the first retrieval information corresponding to the query data can be determined according to the keyword retrieval result, and the second retrieval information corresponding to the query data can be determined according to the vector retrieval result, wherein the first retrieval information refers to the The data related to the query text is selected from the candidate set, and the second retrieval information refers to selecting data related to the query vector from the candidate set to be matched.

It can be understood that after the first search information and the second search information are obtained, the target search information can be generated according to the first search information and the second search information, and then the information query result can be generated according to the target search information.

It can be understood that, in order to improve the generation efficiency of the target retrieval information, the first retrieval information and the second retrieval information can be combined to determine the set of the first retrieval information and the second retrieval information, and then according to the first retrieval information and the second retrieval information The collection of the second retrieval information generates the target retrieval information, that is, the target retrieval information includes the retrieval information obtained by the first retrieval information and the second retrieval information.

In a specific implementation, for example, assuming that the first search information determined through keyword search includes information A, information B, and information C, and assuming that the second search information determined through vector search includes information D and information E, the target search information can be Including information A, information B, information C, information D and information E.

It can be understood that after the information query result is determined, the information query result can be displayed in the query interface, so that the user can easily know the information query result, and can provide the user with generalized query results to meet the user's query needs.

Furthermore, after the generalized query is performed, there may be some data in the query result that is far from the query data, because, in order to avoid this situation, the accuracy of the information query result is improved on the basis of the generalized query, After the step S40, it also includes:

Perform semantic correlation analysis on the information query results through a preset deep learning interactivity model; detect semantic consistency according to the semantic correlation analysis results; obtain target information query results according to the detection results, and perform target information query results on the target information query results. exhibit.

It is understandable that after the information query results are obtained through the above methods, the semantic correlation analysis of the information query results can be performed through the preset deep learning interactive model, and then the semantics of each data in the information query results can be judged according to the semantic correlation analysis results. Consistency, if the semantic consistency check is passed, the information query result will be directly sent. If the semantic consistency check fails, remove some data with large semantic differences in the information query result to obtain the target information query result, or return to the initial step for information query, which is not limited in this embodiment.

It can be understood that after the target information query result is determined, the target information query result can be displayed on the query interface, so that the user can easily know the target information query result. You can refer to Figure 3, which is the overall flow chart of generalized query. This solution uses query text and query vector to use the search engine to query in the previously established index, and uses the deep learning interactive model to judge the semantics according to the results returned by the query. Consistency, the generalization and accuracy of the matching results obtained from this will have a good effect.

In a specific implementation, for example, assuming that the query data is "how to open a treasure chest surrounded by wind in XX game", and there is data in the candidate set to be matched as "a treasure chest surrounded by wind in XX game", the two texts In the original rule matching scenario, it is very difficult to recall and satisfy, and it is even difficult to know that there is data that can be satisfied in the data to be matched in the candidate set to be matched. Open" can retrieve "the treasure chest surrounded by a circle of wind in the XX game", and the calculation semantics is very relevant, which can better meet the needs of users.

It should be noted that this solution is based on scenarios that require precise matching and recall of some special results, and is optimized from the idea of exact matching + rule matching to the idea of retrieval + semantic calculation. It uses keyword retrieval + vector retrieval and cooperates with the semantic model to realize online The precise synonymous matching can maximize the use of data and is applicable to all similar scenarios, which greatly improves the online effect.

It is understandable that this solution can be applied to the onebox recall system of search engines. After using this generalization method, the top3 search results can recall more oneboxes by 3.5%, with a net profit of more than 1%. Key types such as featured abstracts13 % of the recall query comes from this generalized query method, which greatly improves the generalized query effect.

Further, in order to achieve a better semantic correlation analysis effect and make the target information query result more accurate, before performing the semantic correlation analysis on the information query result through the preset deep learning interactivity model, it also includes:

Acquiring second training data; training the initial deep learning model according to the second training data to obtain a preset deep learning interactive model.

It should be understood that the second training data may be sample training data related to semantics, and the training data may be mined to train the initial deep learning model to obtain a deep learning interactive model with a semantic correlation analysis function.

In this embodiment, when a query instruction is received, the query data is determined according to the query instruction; the query text and the query vector are determined according to the query data; keyword retrieval is performed according to the query text, and according to the query vector Carry out vector retrieval; combine keyword retrieval results and vector retrieval results to determine information query results. This scheme uses keyword retrieval and vector retrieval to obtain the retrieval results respectively, and then synthesizes the retrieval results to obtain the information query results, which can generalize the query results and achieve the purpose of generalized query, so as to make full use of the data and more efficiently query the information Carry out generalization, improve the effect of information query, and better meet the needs of users.

In one embodiment, as shown in FIG. 4 , a second embodiment of the information query method of the present invention is proposed based on the first embodiment, and the step S30 includes:

Step S301, perform keyword retrieval according to the query text and a preset keyword index, and perform vector retrieval according to the query vector and a preset vector index.

It should be understood that the preset keyword index and preset vector index can be generated in advance, a database can be built according to the preset keyword index and preset vector index, and then the search engine can be configured. In the process of information query, after the query text and query vector are determined, the query text and query vector can be input into the search engine, and the query text can be searched for keywords through the preset keyword index in the search engine, and at the same time, the search engine can The preset vector index in performs vector retrieval on the query vector.

Among them, there is no fixed order between keyword retrieval and vector retrieval in this embodiment, keyword retrieval and vector retrieval can be performed at the same time, keyword retrieval can be performed first and then vector retrieval, or vector retrieval can be performed first Then perform keyword search, which is not limited in this embodiment.

Further, in order to more efficiently generate the preset keyword index and preset vector index and meet the requirements of keyword matching and vector matching, before the step S301, it also includes:

Acquiring sample data of various business types; generating a candidate set to be matched according to the sample data; generating a preset keyword index and a preset vector index according to the candidate set to be matched.

It should be understood that the sample data of various business types can be obtained, the candidate sets to be matched are generated according to the sample data of these various business types, and then the offline vectors corresponding to the candidate sets to be matched are generated through a preset deep learning representation model. Then generate a preset keyword index according to the candidate set to be matched, and generate a preset vector index according to the offline vector, and then establish a search engine to call the preset keyword index and the preset vector index.

Further, in order to achieve a better vectorization effect, before generating the offline vector corresponding to the candidate set to be matched through the preset deep learning representation model, it also includes:

Acquiring first training data; training an initial deep learning model according to the first training data to obtain a preset deep learning representation model.

It should be understood that the first training data may be sample training data related to vectors, and the training data may be mined to train the initial deep learning model to obtain a vectorized functional deep learning representation model. Wherein, the first training data and the second training data may be the same or different, or partly the same and partly different, which is not limited in this embodiment.

In this embodiment, the preset keyword index and the preset vector index are generated in advance, and the search engine is configured to perform keyword retrieval according to the query text and the preset keyword index, and according to the query vector and the preset vector Index for vector retrieval. In this way, text indexes and vector indexes are established according to the candidate sets to be matched, and complete matching and rule matching are converted into retrieval methods. When a scene that needs to be matched appears, the query data is used to search in the built system, and the idea of a search engine is used Perform query generalization.

In one embodiment, as shown in FIG. 5, a third embodiment of the information query method of the present invention is proposed based on the first embodiment or the second embodiment. In this embodiment, description is made based on the first embodiment. The steps S20, including:

Step S201, performing demand identification processing and vectorization processing respectively according to the query data.

It should be understood that, after determining the query data corresponding to the query instruction, in order to determine the query text and the query vector, demand recognition processing and vectorization processing can be performed according to the query data, and then the query text and query vector can be respectively determined according to the two processing results. query vector.

It can be understood that, in order to achieve better processing effect, the query data can be identified and processed through the requirement identification technology, and the query data can be vectorized through the preset deep learning representation model. Among them, the preset deep learning representation model can have the function of vectorization. Vectorization refers to converting text into vectors. After the query data is determined, the query data can be vectorized through the preset deep learning representation model, so that Get the query vector. Requirement recognition technology can have the function of text recognition and text standardization, can set up and adopt relevant demand recognition technology according to the demand of actual situation, this embodiment does not limit to this, after confirming query data, can use demand recognition technology to query The data is subjected to text recognition and text standardization to obtain the query text.

In step S202, the query text is determined according to the requirement identification processing result, and the query vector is determined according to the vectorization processing result.

It can be understood that, after the requirement recognition processing and vectorization processing are performed on the query data respectively, the query text can be determined according to the requirement recognition processing result, and the query vector can also be determined according to the vectorization processing result, and then according to the query text and the predicted Set the keyword index for keyword retrieval, and perform vector retrieval according to the query vector and the preset vector index to obtain the keyword retrieval result and the vector retrieval result, and then combine the first retrieval information corresponding to the keyword retrieval result and the vector retrieval result corresponding to The second retrieval information is used to generate target retrieval information, and an information query result is generated according to the target retrieval information. In order to improve the semantic correlation and ensure the accuracy of the query while realizing the generalization of the query, the semantic correlation analysis can be carried out on the basis of the information query results to obtain the target information query results, which is suitable for all text matches that require precise determination The scene can make full use of the data, and perform query generalization more efficiently and accurately.

In this embodiment, the requirement recognition processing and vectorization processing are performed respectively according to the query data, the query text is determined according to the requirement recognition processing result, and the query vector is determined according to the vectorization processing result, so that the query text and the query vector can be respectively determined, Then carry out keyword retrieval and vector retrieval at the same time, and cooperate with the semantic model to realize query generalization and improve the accuracy of query results.

In addition, an embodiment of the present invention also proposes a storage medium, on which an information query program is stored, and when the information query program is executed by a processor, the steps of the information query method as described above are implemented.

Since the storage medium adopts all the technical solutions of all the above-mentioned embodiments, it at least has all the beneficial effects brought by the technical solutions of the above-mentioned embodiments, which will not be repeated here.

In addition, referring to FIG. 6 , an embodiment of the present invention also proposes an information query device, the information query device includes:

The query data module 10 is configured to determine query data according to the query instruction when receiving the query instruction.

It should be noted that the information query in this embodiment may include but not limited to text query, and may also include other types of information query, which is not limited in this embodiment. In this embodiment, text query is used as an example for illustration .

It should be noted that the query data in this embodiment refers to the query condition data input by the user, and the query data may include but not limited to text data, and may also include other types of data, which is not limited in this embodiment. In this embodiment, text data is taken as an example for description.

It should be understood that when a user needs to query information, he can input query data in the query interface, and then click a query button to send a query instruction to the computer device. When the computing device receives the query instruction, it can determine the corresponding query data according to the query instruction, and then determine the information query result according to the query data.

In a specific implementation, for example, assuming that the user wants to inquire about the weather tomorrow, the query data is "how is the weather tomorrow" input by the user, and the finally obtained information query result may be "tomorrow will be sunny to cloudy".

A text vector module 20, configured to determine a query text and a query vector according to the query data.

It should be noted that, different from the prior art, this solution determines the two retrieval results in two different ways according to the query text and the query vector, and then combines the two retrieval results to obtain the information query result, so that the information The query results are generalized to obtain more extensive information query results to better meet the needs of users.

It should be understood that requirement recognition processing and vectorization processing may be performed on the query data respectively, so as to obtain query text and query vectors.

It is understandable that when retrieval is required, the preset deep learning representation model can be used to vectorize the query data to obtain a query vector, and at the same time, the query demand recognition technology can be used to perform some standardized processing on the query data to obtain the query text.

The information retrieval module 30 is configured to perform keyword retrieval according to the query text, and perform vector retrieval according to the query vector.

It should be noted that the preset keyword index and preset vector index can be generated in advance according to the candidate set to be matched, and after the query text and query vector are determined in the above-mentioned way, keyword retrieval can be performed according to the query text and preset keyword index , to select the data related to the query text from the candidate set to be matched to obtain the keyword retrieval result. At the same time, vector retrieval can also be performed according to the query vector and the preset vector index, so as to select data related to the query vector from the candidate set to be matched to obtain the vector retrieval result.

The query result module 40 is used to determine the information query result in combination with the keyword search result and the vector search result.

It should be understood that the first retrieval information corresponding to the query data can be determined according to the keyword retrieval result, and the second retrieval information corresponding to the query data can be determined according to the vector retrieval result, wherein the first retrieval information refers to the The data related to the query text is selected from the candidate set, and the second retrieval information refers to selecting data related to the query vector from the candidate set to be matched.

It can be understood that after the first search information and the second search information are obtained, the target search information can be generated according to the first search information and the second search information, and then the information query result can be generated according to the target search information.

It can be understood that, in order to improve the generation efficiency of the target retrieval information, the first retrieval information and the second retrieval information can be combined to determine the set of the first retrieval information and the second retrieval information, and then according to the first retrieval information and the second retrieval information The collection of the second retrieval information generates the target retrieval information, that is, the target retrieval information includes the retrieval information obtained by the first retrieval information and the second retrieval information.

In a specific implementation, for example, assuming that the first search information determined through keyword search includes information A, information B, and information C, and assuming that the second search information determined through vector search includes information D and information E, the target search information can be Including information A, information B, information C, information D and information E.

It can be understood that after the information query result is determined, the information query result can be displayed in the query interface, so that the user can easily know the information query result, and can provide the user with generalized query results to meet the user's query needs.

Furthermore, after the generalized query is performed, there may be some data in the query result that is far from the query data, because, in order to avoid this situation, the accuracy of the information query result is improved on the basis of the generalized query, The information query device also includes a semantic correlation analysis module, which is used to perform semantic correlation analysis on the information query results through a preset deep learning interactivity model; detect semantic consistency according to the semantic correlation analysis results; A target information query result is obtained, and the target information query result is displayed.

It is understandable that after the information query results are obtained through the above methods, the semantic correlation analysis of the information query results can be performed through the preset deep learning interactive model, and then the semantics of each data in the information query results can be judged according to the semantic correlation analysis results. Consistency, if the semantic consistency check is passed, the information query result will be directly sent. If the semantic consistency check fails, remove some data with large semantic differences in the information query result to obtain the target information query result, or return to the initial step for information query, which is not limited in this embodiment.

It can be understood that after the target information query result is determined, the target information query result can be displayed on the query interface, so that the user can easily know the target information query result. You can refer to Figure 3, which is the overall flow chart of generalized query. This solution uses query text and query vector to use the search engine to query in the previously established index, and uses the deep learning interactive model to judge the semantics according to the results returned by the query. Consistency, the generalization and accuracy of the matching results obtained from this will have a good effect.

In a specific implementation, for example, assuming that the query data is "how to open a treasure chest surrounded by wind in XX game", and there is data in the candidate set to be matched as "a treasure chest surrounded by wind in XX game", the two texts In the original rule matching scenario, it is very difficult to recall and satisfy, and it is even difficult to know that there is data that can be satisfied in the data to be matched in the candidate set to be matched. Open" can retrieve "the treasure chest surrounded by a circle of wind in the XX game", and the calculation semantics is very relevant, which can better meet the needs of users.

It should be noted that this solution is based on scenarios that require precise matching and recall of some special results, and is optimized from the idea of exact matching + rule matching to the idea of retrieval + semantic calculation. It uses keyword retrieval + vector retrieval and cooperates with the semantic model to realize online The precise synonymous matching can maximize the use of data and is applicable to all similar scenarios, which greatly improves the online effect.

It is understandable that this solution can be applied to the onebox recall system of search engines. After using this generalization method, the top3 search results can recall more oneboxes by 3.5%, with a net profit of more than 1%. Key types such as featured abstracts13 % of the recall query comes from this generalized query method, which greatly improves the generalized query effect.

Further, in order to achieve a better semantic correlation analysis effect and make the target information query result more accurate, the model training module is also used to obtain the second training data; perform the initial deep learning model according to the second training data Training to get the preset deep learning interactive model.

It should be understood that the second training data may be sample training data related to semantics, and the training data may be mined to train the initial deep learning model to obtain a deep learning interactive model with a semantic correlation analysis function.

In this embodiment, when a query instruction is received, the query data is determined according to the query instruction; the query text and the query vector are determined according to the query data; keyword retrieval is performed according to the query text, and according to the query vector Carry out vector retrieval; combine keyword retrieval results and vector retrieval results to determine information query results. This scheme uses keyword retrieval and vector retrieval to obtain the retrieval results respectively, and then synthesizes the retrieval results to obtain the information query results, which can generalize the query results and achieve the purpose of generalized query, so as to make full use of the data and more efficiently query the information Carry out generalization, improve the effect of information query, and better meet the needs of users.

In an embodiment, the information retrieval module 30 is further configured to perform keyword retrieval according to the query text and a preset keyword index, and perform vector retrieval according to the query vector and a preset vector index.

In one embodiment, the information query device further includes an index generation module, which is used to obtain sample data of various business types; generate a candidate set to be matched according to the sample data; generate a preset key according to the candidate set to be matched Word index and preset vector index.

In an embodiment, the index generating module is further configured to generate an offline vector corresponding to the candidate set to be matched by using a preset deep learning representation model; generate a preset keyword index according to the candidate set to be matched, and A preset vector index is generated according to the offline vector.

In one embodiment, the information query device further includes a model training module, configured to acquire first training data; and train an initial deep learning model according to the first training data to obtain a preset deep learning representation model.

In one embodiment, the information retrieval module 30 is further configured to configure a retrieval engine according to a preset keyword index and a preset vector index; input the query text and the query vector into the retrieval engine; through the The retrieval engine invokes a preset keyword index to perform keyword retrieval on the query text, and the retrieval engine invokes a preset vector index to perform vector retrieval on the query vector.

In one embodiment, the text vector module 20 is further configured to perform requirement recognition processing and vectorization processing according to the query data; determine query text according to the requirement recognition processing result, and determine a query vector according to the vectorization processing result.

In one embodiment, the text vector module 20 is further configured to perform requirement recognition processing on the query data through a requirement recognition technology; perform vectorization processing on the query data through a preset deep learning representation model.

In one embodiment, the query result module 40 is further configured to determine the first search information corresponding to the query data according to the keyword search result; determine the first search information corresponding to the query data according to the vector search result second retrieval information; generating target retrieval information according to the first retrieval information and the second retrieval information; generating information query results according to the target retrieval information.

In one embodiment, the query result module 40 is further configured to combine the first search information and the second search information to determine a set of the first search information and the second search information; Target retrieval information is generated according to a set of the first retrieval information and the second retrieval information.

In one embodiment, the information query device further includes a semantic correlation analysis module, configured to perform semantic correlation analysis on the information query results through a preset deep learning interactivity model; Consistency: obtaining target information query results according to the detection results, and displaying the target information query results.

In one embodiment, the model training module is further used to acquire second training data; and to train an initial deep learning model according to the second training data to obtain a preset deep learning interactive model.

For other embodiments or specific implementation methods of the information query device of the present invention, reference may be made to the above method embodiments, which will not be repeated here.

It should be noted that, in this document, the term "comprising", "comprising" or any other variation thereof is intended to cover a non-exclusive inclusion such that a process, method, article or apparatus comprising a set of elements includes not only those elements, It also includes other elements not expressly listed, or elements inherent in the process, method, article, or device. Without further limitations, an element defined by the phrase "comprising a ..." does not preclude the presence of additional identical elements in the process, method, article, or apparatus comprising that element.

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

Through the description of the above embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus a necessary general-purpose hardware platform, and of course also by hardware, but in many cases the former is better implementation. Based on this understanding, the essence of the technical solution of the present invention or the part that contributes to the prior art can be embodied in the form of a software product, which is stored in a computer-readable storage medium as described above (such as ROM/RAM, magnetic disk, optical disk), including several instructions to make a smart device (which can be a mobile phone, a computing machine, an information query device, or a network information query device, etc.) execute the various embodiments of the present invention. Methods.

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 conversion made by using the description of the present invention and the contents of the accompanying drawings, or directly or indirectly used in other related technical fields , are all included in the scope of patent protection of the present invention in the same way.

The invention discloses A1. An information query method, the information query method comprising:

When receiving a query instruction, determine query data according to the query instruction;

determining a query text and a query vector according to the query data;

performing keyword retrieval according to the query text, and performing vector retrieval according to the query vector;

Combining keyword retrieval results and vector retrieval results to determine information query results.

A2. The information query method as described in A1, wherein the keyword retrieval is performed according to the query text, and the vector retrieval is performed according to the query vector, including:

Keyword retrieval is performed according to the query text and a preset keyword index, and vector retrieval is performed according to the query vector and a preset vector index.

A3, the information query method as described in A2, the keyword retrieval according to the query text and the preset keyword index, and before the vector retrieval according to the query vector and the preset vector index, it also includes:

Obtain sample data of various business types;

generating a candidate set to be matched according to the sample data;

Generate a preset keyword index and a preset vector index according to the candidate set to be matched.

A4. The information query method as described in A3, said generating a preset keyword index and a preset vector index according to the candidate set to be matched, including:

Generate the offline vector corresponding to the candidate set to be matched by preset deep learning representation model;

A preset keyword index is generated according to the candidate set to be matched, and a preset vector index is generated according to the offline vector.

A5, the information query method as described in A4, before generating the offline vector corresponding to the candidate set to be matched through the preset deep learning representation model, it also includes:

Obtain the first training data;

The initial deep learning model is trained according to the first training data to obtain a preset deep learning representation model.

A6. The information query method as described in A2, wherein the keyword retrieval is performed according to the query text and the preset keyword index, and the vector retrieval is performed according to the query vector and the preset vector index, including:

Configure the search engine according to the preset keyword index and preset vector index;

inputting said query text and said query vector into said search engine;

The search engine calls a preset keyword index to perform keyword search on the query text, and the search engine calls a preset vector index to perform vector search on the query vector.

A7. The information query method as described in any one of A1 to A6, said determining query text and query vectors according to said query data, including:

Respectively performing demand identification processing and vectorization processing according to the query data;

The query text is determined according to the requirement identification processing result, and the query vector is determined according to the vectorization processing result.

A8, the information query method as described in A7, said performing demand recognition processing and vectorization processing according to said query data respectively, including:

performing demand identification processing on the query data through a demand identification technology;

The query data is vectorized through a preset deep learning representation model.

A9. The information query method as described in any one of A1 to A6, said combination of keyword retrieval results and vector retrieval results to determine information query results, including:

determining first retrieval information corresponding to the query data according to the keyword retrieval results;

determining second retrieval information corresponding to the query data according to the vector retrieval result;

generating target retrieval information according to the first retrieval information and the second retrieval information;

An information query result is generated according to the target retrieval information.

A10. The information query method as described in A9, said generating target retrieval information according to said first retrieval information and said second retrieval information, comprising:

Combining the first search information and the second search information to determine a set of the first search information and the second search information;

Target retrieval information is generated according to a set of the first retrieval information and the second retrieval information.

A11. The information query method as described in any one of A1 to A6, after determining the information query result in combination with the keyword retrieval result and the vector retrieval result, it also includes:

Performing a semantic correlation analysis on the information query results through a preset deep learning interactivity model;

Detect semantic consistency based on semantic correlation analysis results;

The target information query result is obtained according to the detection result, and the target information query result is displayed.

A12, the information query method as described in A11, before performing semantic correlation analysis on the information query results through the preset deep learning interactive model, it also includes:

Obtain second training data;

The initial deep learning model is trained according to the second training data to obtain a preset deep learning interactive model.

The present invention also discloses B13, an information query device, the information query device includes:

A query data module, configured to determine query data according to the query command when the query command is received;

A text vector module, configured to determine query text and query vectors according to the query data;

An information retrieval module, configured to perform keyword retrieval according to the query text, and perform vector retrieval according to the query vector;

The query result module is used to determine the information query result in combination with the keyword retrieval result and the vector retrieval result.

B14, the information query device as described in B13, the information retrieval module is also used for keyword retrieval according to the query text and preset keyword index, and vector retrieval according to the query vector and preset vector index .

B15, the information query device as described in B14, the information query device also includes:

The index generation module is used to obtain sample data of various business types; generate a candidate set to be matched according to the sample data; generate a preset keyword index and a preset vector index according to the candidate set to be matched.

B16, the information query device as described in B15, the index generation module is also used to generate the offline vector corresponding to the candidate set to be matched through the preset deep learning representation model; generate a preset according to the candidate set to be matched keyword index, and generate a preset vector index according to the offline vector.

B17, the information query device as described in B16, the information query device also includes:

The model training module is used to acquire first training data; train the initial deep learning model according to the first training data to obtain a preset deep learning representation model.

B18, the information query device as described in B14, the information retrieval module is also used to configure a retrieval engine according to a preset keyword index and a preset vector index; input the query text and the query vector into the retrieval engine ; using the search engine to call a preset keyword index to perform keyword search on the query text, and use the search engine to call a preset vector index to perform vector search on the query vector.

The present invention also discloses C19, an information query device. The information query device includes: a memory, a processor, and an information query program stored in the memory and operable on the processor. The information query program When executed by the processor, the above-mentioned information query method is realized.

The present invention also discloses D20, a storage medium, where an information query program is stored on the storage medium, and when the information query program is executed by a processor, the above-mentioned information query method is realized.

Claims (10)

1. An information query method, characterized in that the information query method comprises:

when a query instruction is received, determining query data according to the query instruction;

determining a query text and a query vector according to the query data;

keyword retrieval is carried out according to the query text, and vector retrieval is carried out according to the query vector;

and determining an information query result by combining the keyword search result and the vector search result.

2. The information query method as claimed in claim 1, wherein said performing keyword search based on said query text and performing vector search based on said query vector comprises:

and carrying out keyword retrieval according to the query text and a preset keyword index, and carrying out vector retrieval according to the query vector and a preset vector index.

3. The information query method of claim 2, wherein before the keyword search is performed according to the query text and a preset keyword index, and the vector search is performed according to the query vector and a preset vector index, further comprising:

Acquiring sample data of various service types;

generating a candidate set to be matched according to the sample data;

and generating a preset keyword index and a preset vector index according to the candidate set to be matched.

4. The information query method of claim 3, wherein the generating a preset keyword index and a preset vector index according to the candidate set to be matched comprises:

generating an offline vector corresponding to the candidate set to be matched through a preset deep learning representation model;

and generating a preset keyword index according to the candidate set to be matched, and generating a preset vector index according to the offline vector.

5. The information query method of claim 4, wherein before generating the offline vector corresponding to the candidate set to be matched through the preset deep learning representation model, further comprises:

acquiring first training data;

and training the initial deep learning model according to the first training data to obtain a preset deep learning representation model.

6. The information query method of claim 2, wherein the performing keyword retrieval according to the query text and a preset keyword index and performing vector retrieval according to the query vector and a preset vector index comprises:

Configuring a search engine according to a preset keyword index and a preset vector index;

inputting the query text and the query vector into the search engine;

and calling a preset keyword index through the search engine to perform keyword search on the query text, and calling a preset vector index through the search engine to perform vector search on the query vector.

7. The information query method of any one of claims 1 to 6, wherein said determining query text and query vectors from said query data comprises:

carrying out demand identification processing and vectorization processing according to the query data respectively;

and determining a query text according to the demand recognition processing result, and determining a query vector according to the vectorization processing result.

8. An information inquiry apparatus, characterized in that the information inquiry apparatus includes:

the query data module is used for determining query data according to the query instruction when the query instruction is received;

the text vector module is used for determining a query text and a query vector according to the query data;

the information retrieval module is used for carrying out keyword retrieval according to the query text and carrying out vector retrieval according to the query vector;

And the query result module is used for determining information query results by combining the keyword search results and the vector search results.

9. An information inquiry apparatus, characterized in that the information inquiry apparatus includes: memory, a processor and an information query program stored on the memory and executable on the processor, which when executed by the processor implements the information query method of any one of claims 1 to 7.

10. A storage medium having stored thereon an information inquiry program which, when executed by a processor, implements the information inquiry method according to any one of claims 1 to 7.

Images