CN116992053A - File query method, device, electronic equipment and storage medium - Google Patents

Abstract

The disclosure provides a file query method, relates to the technical field of artificial intelligence, and particularly relates to the technical fields of machine learning, deep learning and natural language processing. The specific implementation scheme is as follows: extracting the sub-text belonging to each attribute from the query text according to the plurality of attribute information to obtain a plurality of query items; recall M initial files according to the multiple query terms, wherein M is an integer greater than 1; determining file characteristics according to interaction behavior information associated with each initial file; and determining the query characteristics of each of the plurality of query terms, and determining the target file from the M initial files according to the query characteristics and the file characteristics. The disclosure also provides a file querying device, an electronic device and a storage medium.

Description

File query method, device, electronic equipment and storage medium

Technical Field

The present disclosure relates to the field of artificial intelligence, and in particular, to the field of machine learning, deep learning, and natural language processing. More particularly, the disclosure provides a file query method, a device, an electronic device and a storage medium.

Background

Files are used as important knowledge carriers, and people often need to find a certain file in work and life. Traditional file finding methods may search based on the topic or content of the file. However, in the case where the user forgets the subject of the file and does not know the content of the file, it is difficult to find a desired file.

Disclosure of Invention

The disclosure provides a file query method, a device, equipment and a storage medium.

According to a first aspect, there is provided a method of querying a file, the method comprising: extracting the sub-text belonging to each attribute from the query text according to the plurality of attribute information to obtain a plurality of query items; recall M initial files according to the multiple query terms, wherein M is an integer greater than 1; determining file characteristics according to interaction behavior information associated with each initial file; and determining the query characteristics of each of the plurality of query terms, and determining the target file from the M initial files according to the query characteristics and the file characteristics.

According to a second aspect, there is provided a document querying device, the device comprising: the extraction module is used for extracting the sub-text belonging to each attribute from the query text according to the plurality of attribute information to obtain a plurality of query items; the recall module is used for recalling M initial files according to a plurality of query items, wherein M is an integer greater than 1; the file feature determining module is used for determining file features according to interaction behavior information associated with each initial file; and the target file determining module is used for determining the query characteristics of each of the plurality of query terms and determining target files from M initial files according to the query characteristics and the file characteristics.

According to a third aspect, there is provided an electronic device comprising: at least one processor; and a memory communicatively coupled to the at least one processor; wherein the memory stores instructions executable by the at least one processor to enable the at least one processor to perform a method provided in accordance with the present disclosure.

According to a fourth aspect, there is provided a non-transitory computer readable storage medium storing computer instructions for causing a computer to perform a method provided according to the present disclosure.

According to a fifth aspect, there is provided a computer program product comprising a computer program stored on at least one of a readable storage medium and an electronic device, which, when executed by a processor, implements a method provided according to the present disclosure.

It should be understood that the description in this section is not intended to identify key or critical features of the embodiments of the disclosure, nor is it intended to be used to limit the scope of the disclosure. Other features of the present disclosure will become apparent from the following specification.

Drawings

The drawings are for a better understanding of the present solution and are not to be construed as limiting the present disclosure. Wherein:

FIG. 1 is a schematic diagram of an exemplary system architecture to which the file querying method and apparatus may be applied, according to one embodiment of the present disclosure;

FIG. 2 is a flow chart of a method of querying a file according to one embodiment of the present disclosure;

FIG. 3 is a schematic diagram of a multi-pass recall method according to one embodiment of the present disclosure;

FIG. 4 is a schematic illustration of computing intersection features between query features and file features according to one embodiment of the present disclosure;

FIG. 5 is an overall framework diagram of a file querying method according to one embodiment of the present disclosure;

FIG. 6 is a block diagram of a file querying device according to one embodiment of the present disclosure;

fig. 7 is a block diagram of an electronic device of a file querying method according to one embodiment of the present disclosure.

Detailed Description

Exemplary embodiments of the present disclosure are described below in conjunction with the accompanying drawings, which include various details of the embodiments of the present disclosure to facilitate understanding, and should be considered as merely exemplary. Accordingly, one of ordinary skill in the art will recognize that various changes and modifications of the embodiments described herein can be made without departing from the scope and spirit of the present disclosure. Also, descriptions of well-known functions and constructions are omitted in the following description for clarity and conciseness.

The files may include documents, images, audio-video, executable programs, netpage cards, and the like. Wherein the netpage card may comprise a IM (Instant Message) card in a chat interface of instant messaging software. In work or life, a user often forgets the file name, so that the file cannot be found. For example, the user remembers only the file that was sent by XX, but does not remember a specific file name. In this case, the conventional search method based on the file title and the content has difficulty in finding a file desired by the user because there is no matching of the file title and the content. In addition, conventional file searches often do not support complex search instructions expressed in natural language (e.g., find XX to My files).

Especially in an office scene, the files are numerous, a user often cannot view each file in time, and the situation that the user needs to look for the files in a round is common. If the file name is forgotten, it is difficult to find a desired file. This results in problems of low file searching efficiency, long file searching path, and high file screening cost.

There are also rule-based file searching methods, semantic-based file searching methods, and the like in the related art. Wherein the rule-based file search method may use predefined rules to match relationships between query statements and files. Rules may be defined based on specific vocabulary, structure, or other criteria. However, defining rules is labor-intensive, and one rule can only solve one problem with limited effectiveness. As for the semantic-based file search method, although related files can be matched by understanding user intention, a large amount of high-quality training data is required to achieve a precise semantic search effect. In addition, semantic-based search results are also poor for complex search instructions expressed in natural language.

In the technical scheme of the disclosure, the related processes of collecting, storing, using, processing, transmitting, providing, disclosing and the like of the personal information of the user accord with the regulations of related laws and regulations, and the public order colloquial is not violated.

In the technical scheme of the disclosure, the authorization or consent of the user is obtained before the personal information of the user is obtained or acquired.

FIG. 1 is a schematic diagram of an exemplary system architecture to which the method and apparatus for querying a file may be applied, according to one embodiment of the present disclosure. It should be noted that fig. 1 is only an example of a system architecture to which embodiments of the present disclosure may be applied to assist those skilled in the art in understanding the technical content of the present disclosure, but does not mean that embodiments of the present disclosure may not be used in other devices, systems, environments, or scenarios.

As shown in fig. 1, a system architecture 100 according to this embodiment may include terminal devices 101, 102, 103, a network 104, and a server 105. The network 104 is used as a medium to provide communication links between the terminal devices 101, 102, 103 and the server 105. The network 104 may include various connection types, such as wired and/or wireless communication links, and the like.

The user may interact with the server 105 via the network 104 using the terminal devices 101, 102, 103 to receive or send messages or the like. The terminal devices 101, 102, 103 may be a variety of electronic devices including, but not limited to, smartphones, tablets, laptop portable computers, and the like.

The file querying method provided by the embodiments of the present disclosure may be generally performed by the server 105. Accordingly, the file querying apparatus provided in the embodiments of the present disclosure may be generally disposed in the server 105. The file querying method provided by the embodiments of the present disclosure may also be performed by a server or a server cluster that is different from the server 105 and is capable of communicating with the terminal devices 101, 102, 103 and/or the server 105. Accordingly, the file querying apparatus provided by the embodiments of the present disclosure may also be provided in a server or a server cluster that is different from the server 105 and is capable of communicating with the terminal devices 101, 102, 103 and/or the server 105.

Fig. 2 is a flow chart of a method of querying a file according to one embodiment of the present disclosure.

As shown in fig. 2, the file querying method 200 includes operations S210 to S240.

In operation S210, according to the plurality of attribute information, sub-texts belonging to the respective attributes are extracted from the query text, thereby obtaining a plurality of query terms.

In searching for files, if the user forgets the file name and does not know the content of the file, the user would prefer to be able to find the document they want by way of natural language expression. For example, "find XX yesterday shared with My meeting summary documents in a small team group". The query text may be the query instruction expressed in natural language.

For query instructions expressed in natural language, the query instructions can be analyzed according to a plurality of preset slots. For example, each slot has attribute information, which may include "person name", "action", "time", "subject", "scene", "file type", and the like. Sub-texts belonging to each attribute can be extracted from the query text, and the sub-texts are filled into corresponding slots, so that a plurality of query items are obtained.

For example, the query text described above may be parsed into a plurality of query terms: name of person: XX "," time: yesterday "," behaviour: sharing "," topic: meeting summary "," scene: small teams group "," file type: a document).

The embodiment can provide a file retrieval system, through the query terms, the file retrieval system can know that a user wants to find a document which is shared by XX, the time of the sharing action is yesterday, the sharing scene is in a small team group, and if the document is also related to a meeting summary, the document should be ranked in front.

It should be noted that if the query text does not include some attribute information, the corresponding query term is empty. For example, if no time information is contained in the query text, then the query term for the time attribute may be null.

In operation S220, M initial files are recalled according to the plurality of query terms.

For example, a plurality of query terms may be entered into a file matching model, which may recall initial files related to the plurality of query terms from a database. Wherein the file matching model may be a trained deep learning model. Further, the file matching model may be modeled based on the interaction behavior information of the file. The interactive behavior information of the file refers to the interactive behavior of the user on the file, and the interactive behavior can comprise display/screen projection in a meeting, sharing in an IM group chat/private chat interface, creating, editing and browsing the file in a knowledge base and the like.

For example, the file matching model, during the training phase, the input of the model may include at least query terms for behavioral attributes. It should be noted that, in addition to the query term of the behavior attribute, the input of the model may further include at least one of a query term of the name attribute, a query term of the time attribute, a query term of the topic attribute, a query term of the scene attribute, and a query term of the type attribute. For example, the input to the model may be a file created by XX, a file edited by XX yesterday, a file shared by XX in a small team group, and so on. The output of the model may be the file most relevant to the input information as determined by the model. And then determining loss according to the difference between the file actually wanted by the user and the file output by the model, and further adjusting parameters of the model according to the loss until the model converges.

Thus, the trained file matching model is capable of recalling relevant initial files from the database based on the file interaction behavior information.

In operation S230, for each initial file, file characteristics are determined according to the interaction behavior information associated with the initial file.

The interaction behavior information associated with the initial file may include interaction behavior information of the user with respect to the initial file. Including, for example, interactive objects (e.g., user's name), interactive behavior patterns (e.g., sharing), interaction time, interaction scenario (e.g., team group), interaction topic (e.g., meeting summary), and file type (e.g., document, picture, etc.).

For each initial file, at least one of an interaction behavior pattern feature, an interaction object feature, an interaction time feature, an interaction scene feature, and an interaction topic feature of the initial file may be determined as a file feature of the initial file.

For example, the interactive behavior information may be input into a convolutional network, and text features of each interactive behavior information may be extracted as file features. Or inputting the interactive behavior information into a natural language processing model to obtain semantic features of each interactive behavior information as file features.

It should be noted that, for each initial file, if the initial file has a plurality of interaction information, a plurality of file features may be obtained. For example, file features corresponding to sharing behaviors, file features corresponding to creation behaviors, file features corresponding to editing behaviors, and so forth may be included.

In operation S240, query characteristics of each of the plurality of query terms are determined, and a target file is determined from the M initial files according to the query characteristics and the file characteristics.

Corresponding to the interaction behavior information, the plurality of attribute information of the query text may include a target behavior manner (e.g., sharing, creating, editing, etc.), a target object (e.g., a user's name), a target time, a target scene (e.g., a team group), a target topic (e.g., a meeting summary), and a target type (e.g., a document, a picture, etc.). Correspondingly, the plurality of query terms includes a target behavior style query term, a target object query term, a target time query term, a target scenario query term, a target topic query term, and a target type query term.

The target behavior pattern feature, the target object feature, the target time feature, the target scene feature, the target subject feature, and the target type feature may be determined as query features for each of the plurality of query terms.

For example, a plurality of query terms may be input into a convolutional network, and text features of each query term may be extracted as query features. Or inputting a plurality of query terms into a natural language processing model and the like to obtain semantic features of each query term information as query features.

Next, a target file is determined from the M initial files based on the query features and the text features.

For example, a correlation (e.g., similarity) between the query feature and the file feature of each initial file may be calculated, with the initial file having the largest correlation being the target file.

The relevance can also be used as a feature of one dimension, the feature of the query and the feature of the file are input into a sorting model together, the respective evaluation values of the M initial files are obtained, sorting is carried out according to the evaluation values, and the initial file with the highest evaluation value is determined to be used as the target file. The ranking model may be a machine learning model, such as XGBoost (eXtreme Gradient Boosting, machine learning algorithm of extreme gradient lifted trees), random forest model, and the like.

According to the embodiment of the disclosure, a plurality of query items are extracted from a query text according to a plurality of attribute information, M initial files are recalled by using the plurality of query items, file characteristics are determined according to interaction behavior information with the initial files, and target files are determined according to the query characteristics and the file characteristics. According to the method and the device, the corresponding file can be found according to the interactive behavior information of the file, the query instruction expressed in the natural language is supported, and the file finding efficiency is improved.

It can be understood that, in the case that the user forgets the document theme or the unknown document content, the embodiment can find the desired document through the search instruction which is expressed in the natural language and contains the document interaction behavior information, so that a more accurate and personalized document search result can be provided, and the search experience of the user is improved.

According to an embodiment of the present disclosure, operation S210 includes inputting a query text, a plurality of attribute information, and query constraint information into a large language model to obtain a plurality of query terms, where the query constraint information is used to instruct the large language model to extract sub-texts belonging to respective attributes from the query text, and determining the plurality of attribute information and the sub-texts of the respective plurality of attribute information as a plurality of query terms.

For example, a large language model (Large Language Model, LLM) may be utilized to parse query text into a plurality of query terms. The query text, the preset plurality of slot information and the query constraint information can be input into a large language model to obtain a plurality of query items. The slot information includes attribute information, that is, each slot has attribute information, and the attribute information may include "person name", "action", "time", "theme", "scene", "file type", and the like. The query constraint information is convention information related to the input and output of the model, the convention information is used for indicating the large language model to extract the sub-text belonging to each slot from the query text according to preset slot information, and the sub-text is filled into the corresponding slot to obtain the query item.

Because the large language model has strong natural language understanding capability, the query text is analyzed into specific query terms by using the large language model, and the search intention of the user can be accurately understood and positioned, so that the understanding capability of complex search instructions expressed in natural language can be improved.

The embodiment utilizes the large voice model and the preset slot position information, so that the document retrieval system can better understand and meet the search requirement of a user. By analyzing the search instruction of the user into specific query term information, more accurate and personalized file search results can be provided, and the search experience of the user is improved.

According to an embodiment of the present disclosure, operation S220 may recall M initial files in a multi-pass recall manner.

Fig. 3 is a schematic diagram of a multi-pass recall method according to one embodiment of the present disclosure.

As shown in fig. 3, the plurality of recall paths of the present embodiment include an interactive behavior based recall path 310, a personally recall path 320, and a topic recall path 330.

For the recall path 310 based on interaction behavior, the present embodiment recalls at least one first initial file from the database according to the target object query term, the target behavior pattern query term, the target time query term, the target scene query term, the target topic query term, and the target type query term. The database may include a full amount of file resources including, for example, meeting use/show/screen files, IM group chat shared files, IM private chat shared files, and files created/edited/browsed in a knowledge base, etc.

The interaction-based recall channel 310 may include a first file matching model whose input may include a plurality of query terms and whose output may be at least one first initial file recalled from the database that is most relevant to the plurality of query terms.

The first file matching model may be modeled based on the interaction behavior information of the file. For example, the first file matching model, during the training phase, the input of the model may include at least query terms for behavioral attributes. It should be noted that, in addition to the query term of the behavior attribute, the input of the model may further include at least one of a query term of the name attribute, a query term of the time attribute, a query term of the topic attribute, a query term of the scene attribute, and a query term of the type attribute. Thus, the trained first file matching model is capable of recalling relevant initial files from the database based on the file interaction behavior information.

The target object query term, the target behavior mode query term, the target time query term, the target scene query term, the target subject query term and the target type query term can be input into a trained first file matching model to obtain at least one first initial file.

For the person-around recall path 320, the embodiment recalls at least one second initial file from the preset file set according to the target object query term, the target behavior mode query term, the target time query term, the target scene query term, the target subject query term and the target type query term, wherein the interaction behavior information of the initial files in the preset file set is consistent with at least one of the query terms.

The persona recall path 320 may include a second file matching model whose input may include a plurality of query terms and whose output may be recall from a set of preset files at least one second initial file most relevant to the plurality of query terms. The bystander recall path 320 may include a second file matching model that may also be modeled based on file interaction behavior information. The target object query term, the target behavior mode query term, the target time query term, the target scene query term, the target subject query term and the target type query term can be input into a trained second file matching model to obtain at least one second initial file.

The person-in-person recall path 320 differs from the interaction behavior-based recall path 310 in that the interaction behavior-based recall path 310 recalls an initial file from a database and the person-in-person recall path 320 recalls an initial file from a set of pre-set files. The database may contain a full initial set of files. And the preset file set may be an initial file set meeting a predetermined condition, and the preset file set may be a set of partial files in the database.

The predetermined condition may include that interaction behavior information of an initial file in the preset file set is consistent with the target object query term. For example, the target object (name) of the target object query belongs to a certain group, and the initial files in the preset file set are all files having interaction behaviors with the members (people around) in the group. That is, the initial files obtained by the operations of creating, editing, sharing and the like of the members in the group are stored in the preset file set, and the initial files having interaction with the nearby person can be recalled through the nearby person recall path 320.

The purpose of setting the person-to-person recall channel 320 is that the person-to-person recall channel 320 can be used as a spam scheme to return an initial file with interaction with the person-to-person for the user in the case that the query item does not contain the target object, so that the condition that the recall file is empty is avoided.

It should be noted that, the predetermined condition may further include that the interaction behavior information of the initial file in the preset file set is consistent with other query terms (at least one of the target behavior mode feature, the target time feature, the target scene feature and the target theme feature). Similar to the person-in-person recall path 320, the search range of the initial file is limited to a certain range, and at least the initial file within the range can be returned to the user. For example, the interaction scene in the preset file set is consistent with the target scene in the query term, and the initial file under the target scene (e.g., the small team) can be returned.

For the topic recall path 330, the present embodiment recalls at least one third initial file from the database based on the target topic query term.

The topic recall path 330 can include a third file matching model whose input can include a target topic query term and whose output can be recall from the database at least one third initial file that is most relevant to the target topic query term.

The third file matching model may be modeled based on the topic of the file. The theme may include a file name and content. For example, a document, a topic may include a title and content. The third document matching model, during the training phase, the input of the model may include a subject query term. Thus, the trained third document matching model can have the ability to recall relevant initial documents from the database based on the document theme.

The target subject query term may be input into a trained third document matching model to obtain at least one third initial document.

Next, the present embodiment determines at least one first initial file, at least one second initial file, and at least one third initial file as M initial files.

For example, through the interactive behavior based recall path 310, the personals recall path 320, and the topic recall path 330, 100 first initial files, 100 second initial files, and 100 third initial files are recalled, respectively, 300 initial files can be obtained.

According to the embodiment, M initial files are recalled through the recall passages, and recall modes of multiple dimensions such as file interaction behavior information and content topics are combined, so that the files wanted by a user can be ensured to be covered in a recall set, more possible matching results can be provided, and the retrieval requirements of the user can be more comprehensively met.

It will be appreciated that the recall path 310 based on interaction behavior is modeled based on the interaction behavior of the file, and thus, for the case where the user forgets the specific name of the file, the desired file may still be found based on the file interaction behavior (e.g., IM sharing, editing, creation, etc.).

According to an embodiment of the present disclosure, operation S240 includes determining, for each initial file, a correlation between a query feature and a file feature of the initial file as a cross feature; and determining the target file from M initial files according to the query characteristics, the file characteristics and the cross characteristics.

The crossover feature is described below.

FIG. 4 is a schematic diagram of computing intersection features between query features and file features according to one embodiment of the present disclosure.

As shown in fig. 4, query feature 410 includes features of each of a plurality of query terms, for example, the plurality of query terms includes a target behavior pattern (sharing), a target object (person name), a target time (yesterday), a target scene (team group), a target topic (meeting summary), a target type (document).

The file feature 420 may include a plurality of, for example, file features 421 corresponding to the creation behavior, file features 422 corresponding to the sharing behavior, and the like, for each initial file. The file feature 421 includes, among others, an interactive behavior manner (creation), an interactive object (creator), an interactive theme (file title), an interactive time (creation time), and a file type. The file features 422 include interactive behavior patterns (sharing), interactive objects (sharing people), interactive time (sharing time), interactive scene features (sharing scenes).

Correlations between each of the query features 410 and the file features 421, respectively, may be computed resulting in cross features related to the creation behavior. The intersection features may include at least one of behavioral style intersection features, object intersection features, time intersection features, scene intersection features, subject intersection features, and type intersection features. Wherein the correlation may be a similarity or a numerical value characterizing yes or no. For example, the similarity between the target object feature (person name) in the query feature 410 and the interaction object feature (creator) in the file feature 421 is calculated as the object cross feature. For another example, it is determined whether the target time (yesterday) in query feature 410 matches the interaction time feature (creation time) in file feature 421, if so, the time-crossing feature is 1, otherwise the time-crossing feature is 0.

Similarly, the relevance between each of the query features 410 and the file features 422, respectively, may be calculated, resulting in cross-features related to sharing behavior.

Next, the cross-over feature associated with the creation behavior and the cross-over feature associated with the sharing behavior are combined, resulting in an overall cross-over feature 430.

According to an embodiment of the present disclosure, determining a target file from M initial files according to query characteristics, file characteristics, and cross characteristics includes: according to the query term characteristics, the file characteristics and the cross characteristics, determining respective evaluation values of M initial files; determining N candidate files from M initial files according to the evaluation values, wherein N is an integer greater than 1; and determining the target file from the N candidate files according to the correlation between the query text and the N candidate files.

According to the embodiment of the disclosure, the query feature 410, the file feature 420 and the cross feature 430 can be used as the features of multiple dimensions, and the M initial files are ranked by multi-feature fusion, so as to determine the target file.

For example, the XGBoost model may be used as a ranking model for multi-feature fusion. The XGBoost model is high in running speed, and can be used as a coarse ordering model to quickly determine N candidate files from M initial files and serve as input of subsequent fine ordering.

For example, query feature 410, file feature 420 for each initial file, and cross feature 430 between query feature 410 and file feature 420 for each initial file may be input into a trained XGBoost model to obtain an evaluation value for each initial file. The evaluation value characterizes the relevance between the initial files and the query features, the initial files are ranked according to the relevance, and N (for example, the first 20) initial files can be selected from the ranking results as candidate files.

Next, a fine ordering may be performed for the N candidate files to determine the target file.

According to the embodiment of the disclosure, a query text, N candidate files and target constraint information are input into a large language model to obtain target files, wherein the target constraint information is used for indicating the large language model to search out target files most relevant to the query text from the N candidate files.

In the fine ranking stage, a large language model may be utilized to further fine rank among the N candidate files returned by the coarse ranking stage. For example, the query text, the N candidate files, and the target constraint information are input into a large language model, which outputs the target document. The target constraint information may represent contracted information associated with the model input and the output, the contracted information including that the model input is query text and a plurality of candidate documents, the model being required to output a candidate document of the plurality of candidate documents that is most relevant to the query text as the target document.

The correlation between the query text and the candidate files can be more accurately measured by utilizing a large language model, thereby providing finer file ordering results. The large language model may be an SFT (Supervisory fine-tuning) based model.

The fusion ordering method of the embodiment comprises two stages of coarse ordering and fine ordering, wherein the coarse ordering stage can improve the coarse ordering effect of document retrieval while ensuring the retrieval efficiency by constructing multidimensional features and returning candidate files by utilizing an XGBoost model. And the fine sorting stage calculates the relevance between the query text and the candidate file by using the large language model, and returns a finer sorting result. The comprehensive method can ensure the file retrieval efficiency as much as possible and effectively improve the file ordering effect.

Fig. 5 is an overall framework diagram of a file querying method according to one embodiment of the present disclosure.

The present embodiment includes a large language model 510, a Query understanding module 520, a multi-recall module 530, and a fusion ordering module 540.

When the user searches for a file, a search instruction expressed in natural language, for example, "find meeting summary that XX shared in small team group yesterday" is input to the Query understanding module 520. The Query understanding module 520 inputs the Query, the preset plurality of slot information and the Query constraint information into the large language model 510, and analyzes the Query according to a plurality of preset operations by using the large language model 510 to obtain a plurality of Query terms. The respective "objects: XX "," behavior: IM sharing "," time: yesterday "," scene: small teams "," subject: meeting summary). It should be noted that, the slot may also include a file type, and since the Query does not include file type information, the file type slot may be empty.

The multi-way recall module 530 includes interactive behavior based recall pathways, personally-located recall pathways, and theme recall pathways. The recall path based on the interaction is modeled based on the interaction information of the file, and the interaction information comprises meeting display/screen throwing, IM group chat/private chat sharing, creation, editing, browsing and the like. The plurality of query terms are entered into an interactive behavior based recall path that recalls from the database at least one first initial file associated with the plurality of query terms. The database may include a full amount of file resources, including files for meeting use/presentation and screen projection, files for IM group chat sharing, files for IM private chat sharing, files for creation/editing/browsing in a knowledge base, and so on.

The personally recall path may also be modeled based on file-based interaction behavior information. The plurality of query terms are entered into a personally recall path that recalls at least one second initial file associated with the plurality of query terms from a set of preset files. The difference between the personally recalled path and the interactive behavior-based recall path is that the personally recalled path limits the search scope of the initial file to a certain extent, for example, in a preset file collection that has interactive behaviors with personnel in the group. Under the condition that the query item does not contain an object, the nearby person recall passage can be used as a spam scheme to return an initial file with interaction behavior with nearby persons for a user, so that the situation that the recall file is empty is avoided.

The topic recall path can be modeled based on the topic of the file. The theme may include a file name and content. For example, a document, a topic may include a title and content. The subject query term may be input into a trained third file matching model to obtain at least one third initial file.

The at least one first initial file, the at least one second initial file, and the at least one third initial file are determined as M initial files as inputs to the merge sort module 540.

The fused ordering module 540 includes a coarse ordering stage and a fine ordering stage. In the coarse ordering stage, the respective query characteristics of a plurality of query terms are first determined, and the file characteristics of each initial file are determined. Query features include respective features of behavior, objects, time, subject, scene. The file characteristics of each initial file may have a plurality of file characteristics, for example, file characteristics related to the creation behavior and file characteristics related to the sharing behavior, where the file characteristics related to the creation behavior include respective characteristics of creator, file title, creation time, etc., and the file characteristics related to the sharing behavior include respective characteristics of sharer, sharing time, sharing scene, etc.

Then, the relevance between each query feature and each file feature can be calculated to obtain the cross feature. The cross feature may include a similarity or a numerical value indicating yes or no. For example, the cross feature between the object query feature and the sharer (creator) may be a similarity between the two. If the time query feature is consistent with the sharing time (or creation time), the intersection feature of the two is 1, otherwise, the intersection feature is 0. In this way, cross-features between each query feature and the respective file features may be obtained.

And then, the query features, the file features and the cross features are used as the features of multiple dimensions, and the M initial files are subjected to rough sorting of multi-feature fusion to obtain candidate files. For example, the XGBoost model may be used as a coarse ordering model for multi-feature fusion, the query features, the file features and the cross features are input into the XGBoost model to obtain a coarse ordering result, and the first N initial files are selected from the coarse ordering result as candidate files, and the N candidate files are used as inputs for fine ordering.

In the fine sorting stage, a large language model can be used as a fine sorting model, query, N candidate files and target constraint information are input into the large language model, and the large language model outputs a target document.

The method and the device can accurately position the search intention of the user by utilizing the strong natural language understanding capability of the large language model, thereby improving the understanding capability of the file retrieval system on the complex search instruction and enabling the user to retrieve the document in a natural language mode.

Secondly, in the file retrieval process, the embodiment can more comprehensively meet the retrieval requirements of users by using a plurality of recall paths and provide more possible matching results. Wherein the recall path based on the user interaction behavior can still find the required document through the file interaction behavior in case the user forgets the specific name of the file.

In addition, the embodiment adopts a multi-feature fusion coarse ordering and a fine ordering method based on a large language model. The coarse ranking stage is used for more comprehensively evaluating the relevance of the file by combining the characteristic information of multiple dimensions. And in the fine-ranking stage, calculating the correlation between the Query and the file by using a large language model, and returning to the target file. The comprehensive method can ensure the file retrieval efficiency as much as possible and provide more accurate file retrieval results for users.

Fig. 6 is a block diagram of a file querying device according to one embodiment of the present disclosure.

As shown in fig. 6, the file querying apparatus 600 includes an extraction module 601, a recall module 602, a file feature determination module 603, and a target file determination module 604.

The extraction module 601 is configured to extract, according to the plurality of attribute information, sub-texts belonging to each attribute from the query text, so as to obtain a plurality of query terms.

Recall module 602 is configured to recall M initial files based on a plurality of query terms, where M is an integer greater than 1.

The file characteristic determining module 603 is configured to determine, for each initial file, a file characteristic according to interaction behavior information associated with the initial file.

The target file determining module 604 is configured to determine query characteristics of each of the plurality of query terms, and determine a target file from the M initial files according to the query characteristics and the file characteristics.

According to an embodiment of the present disclosure, the interaction behavior information includes at least one of an interaction behavior pattern, an interaction object, an interaction time, an interaction scene, an interaction topic, and a file type.

The file feature determining module 603 is configured to determine at least one of an interaction behavior mode feature, an interaction object feature, an interaction time feature, an interaction scene feature, an interaction topic feature, and a file type feature as a file feature according to the interaction behavior information.

According to an embodiment of the present disclosure, the attribute information includes one of a target behavior pattern, a target object, a target time, a target scene, a target topic, and a target type, and the query term includes one of a target behavior pattern query term, a target object query term, a target time query term, a target scene query term, a target topic query term, and a target type query term.

The target file determination module 604 is further configured to determine a target behavior pattern feature, a target object feature, a target time feature, a target scene feature, a target subject feature, and a target type feature as query features for each of the plurality of query terms.

The object file determination module 604 includes a cross feature determination unit and an object file determination unit.

The cross feature determination unit is used for determining correlation between the query feature and the file feature of the initial file as a cross feature for each initial file.

The target file determining unit is used for determining target files from M initial files according to the query characteristics, the file characteristics and the cross characteristics.

The cross feature determining unit comprises an object cross feature determining subunit, a time cross feature determining subunit, a scene cross feature determining subunit, a theme cross feature determining subunit and a type cross feature determining subunit.

The behavior pattern intersection feature determination subunit is configured to determine a correlation between the target behavior pattern feature and the file feature as a behavior pattern intersection feature.

The object cross feature determination subunit is configured to determine a correlation between the target object feature and the file feature as an object cross feature.

The time-crossing feature determination subunit is configured to determine a correlation between the target time feature and the file feature as a time-crossing feature.

The scene intersection feature determination subunit is configured to determine a correlation between the target scene feature and the file feature as a scene intersection feature.

The theme-cross-feature determination subunit is configured to determine a correlation between the target theme feature and the file feature as a theme-cross feature.

The type-crossing feature determination subunit is configured to determine a correlation between the target type feature and the file feature as a type-crossing feature.

The target file determining unit includes an evaluation value determining subunit, a candidate file determining subunit, and a target file determining subunit.

The evaluation value determination subunit is configured to determine respective evaluation values of the M initial files according to the query term feature, the file feature, and the cross feature.

The candidate file determining subunit is configured to determine N candidate files from the M initial files according to the evaluation values, where N is an integer greater than 1.

The target file determining subunit is configured to determine a target file from the N candidate files according to a correlation between the query text and the N candidate files.

The target file determining subunit is further configured to input the query text, the N candidate files, and target constraint information into the large language model to obtain a target file, where the target constraint information is used to instruct the large language model to find a target file most relevant to the query text from the N candidate files.

Recall module 602 includes a first recall unit, a second recall unit, a third recall unit, and an initial file determination unit.

The first recall unit is used for recalling at least one first initial file from the database according to the plurality of query items.

The second recall unit is used for recalling at least one second initial file from the preset file set according to the plurality of query items, wherein the interaction behavior information of the initial files in the preset file set is consistent with at least one of the plurality of query items.

The third recall unit is used for recalling at least one third initial file from the database according to the target subject query item.

The initial file determining unit is used for determining the first initial file, the second initial file and the third initial file as M initial files.

The extraction module 601 is configured to input a query text, a plurality of attribute information, and query constraint information into a large language model to obtain a plurality of query terms, where the query constraint information is used to instruct the large language model to extract sub-texts belonging to each attribute from the query text, and determine the plurality of attribute information and the sub-texts of each attribute as a plurality of query terms.

According to embodiments of the present disclosure, the present disclosure also provides an electronic device, a readable storage medium and a computer program product.

Fig. 7 illustrates a schematic block diagram of an example electronic device 700 that may be used to implement embodiments of the present disclosure. Electronic devices are intended to represent various forms of digital computers, such as laptops, desktops, workstations, personal digital assistants, servers, blade servers, mainframes, and other appropriate computers. The electronic device may also represent various forms of mobile devices, such as personal digital processing, cellular telephones, smartphones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions, are meant to be exemplary only, and are not meant to limit implementations of the disclosure described and/or claimed herein.

As shown in fig. 7, the apparatus 700 includes a computing unit 701 that can perform various appropriate actions and processes according to a computer program stored in a Read Only Memory (ROM) 702 or a computer program loaded from a storage unit 708 into a Random Access Memory (RAM) 703. In the RAM 703, various programs and data required for the operation of the device 700 may also be stored. The computing unit 701, the ROM 702, and the RAM 703 are connected to each other through a bus 704. An input/output (I/O) interface 705 is also connected to bus 704.

Various components in device 700 are connected to I/O interface 705, including: an input unit 706 such as a keyboard, a mouse, etc.; an output unit 707 such as various types of displays, speakers, and the like; a storage unit 708 such as a magnetic disk, an optical disk, or the like; and a communication unit 709 such as a network card, modem, wireless communication transceiver, etc. The communication unit 709 allows the device 700 to exchange information/data with other devices via a computer network, such as the internet, and/or various telecommunication networks.

The computing unit 701 may be a variety of general and/or special purpose processing components having processing and computing capabilities. Some examples of computing unit 701 include, but are not limited to, a Central Processing Unit (CPU), a Graphics Processing Unit (GPU), various specialized Artificial Intelligence (AI) computing chips, various computing units running machine learning model algorithms, a Digital Signal Processor (DSP), and any suitable processor, controller, microcontroller, etc. The computing unit 701 performs the respective methods and processes described above, such as a file querying method. For example, in some embodiments, the file querying method may be implemented as a computer software program tangibly embodied on a machine-readable medium, such as storage unit 708. In some embodiments, part or all of the computer program may be loaded and/or installed onto device 700 via ROM 702 and/or communication unit 709. When the computer program is loaded into RAM 703 and executed by computing unit 701, one or more steps of the file querying method described above may be performed. Alternatively, in other embodiments, the computing unit 701 may be configured to perform the file querying method by any other suitable means (e.g., by means of firmware).

Various implementations of the systems and techniques described here above may be implemented in digital electronic circuitry, integrated circuit systems, field Programmable Gate Arrays (FPGAs), application Specific Integrated Circuits (ASICs), application Specific Standard Products (ASSPs), systems On Chip (SOCs), complex Programmable Logic Devices (CPLDs), computer hardware, firmware, software, and/or combinations thereof. These various embodiments may include: implemented in one or more computer programs, the one or more computer programs may be executed and/or interpreted on a programmable system including at least one programmable processor, which may be a special purpose or general-purpose programmable processor, that may receive data and instructions from, and transmit data and instructions to, a storage system, at least one input device, and at least one output device.

Program code for carrying out methods of the present disclosure may be written in any combination of one or more programming languages. These program code may be provided to a processor or controller of a general purpose computer, special purpose computer, or other programmable data processing apparatus such that the program code, when executed by the processor or controller, causes the functions/operations specified in the flowchart and/or block diagram to be implemented. The program code may execute entirely on the machine, partly on the machine, as a stand-alone software package, partly on the machine and partly on a remote machine or entirely on the remote machine or server.

In the context of this disclosure, a machine-readable medium may be a tangible medium that can contain, or store a program for use by or in connection with an instruction execution system, apparatus, or device. The machine-readable medium may be a machine-readable signal medium or a machine-readable storage medium. The machine-readable medium may include, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any suitable combination of the foregoing. More specific examples of a machine-readable storage medium would include an electrical connection based on one or more wires, a portable computer diskette, a hard disk, a Random Access Memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

To provide for interaction with a user, the systems and techniques described here can be implemented on a computer having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to a user; and a keyboard and pointing device (e.g., a mouse or trackball) by which a user can provide input to the computer. Other kinds of devices may also be used to provide for interaction with a user; for example, feedback provided to the user may be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user may be received in any form, including acoustic input, speech input, or tactile input.

The systems and techniques described here can be implemented in a computing system that includes a background component (e.g., as a data server), or that includes a middleware component (e.g., an application server), or that includes a front-end component (e.g., a user computer having a graphical user interface or a web browser through which a user can interact with an implementation of the systems and techniques described here), or any combination of such background, middleware, or front-end components. The components of the system can be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include: local Area Networks (LANs), wide Area Networks (WANs), and the internet.

The computer system may include a client and a server. The client and server are typically remote from each other and typically interact through a communication network. The relationship of client and server arises by virtue of computer programs running on the respective computers and having a client-server relationship to each other.

It should be appreciated that various forms of the flows shown above may be used to reorder, add, or delete steps. For example, the steps recited in the present disclosure may be performed in parallel, sequentially, or in a different order, provided that the desired results of the disclosed aspects are achieved, and are not limited herein.

The above detailed description should not be taken as limiting the scope of the present disclosure. It will be apparent to those skilled in the art that various modifications, combinations, sub-combinations and alternatives are possible, depending on design requirements and other factors. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present disclosure are intended to be included within the scope of the present disclosure.

Claims (21)

1. A method of querying a file, comprising:

extracting the sub-text belonging to each attribute from the query text according to the plurality of attribute information to obtain a plurality of query items;

recall M initial files according to the plurality of query terms, wherein M is an integer greater than 1;

determining file characteristics according to interaction behavior information associated with each initial file; and

and determining the query characteristics of each of the plurality of query terms, and determining a target file from the M initial files according to the query characteristics and the file characteristics.

2. The method of claim 1, wherein the interaction behavior information includes at least one of an interaction behavior pattern, an interaction object, an interaction time, an interaction scene, an interaction topic, and a file type; the determining file characteristics according to the interaction behavior information associated with the initial file comprises the following steps of: for each of the initial files,

And determining at least one of an interactive behavior mode feature, an interactive object feature, an interactive time feature, an interactive scene feature, an interactive theme feature and a file type feature as the file feature according to the interactive behavior information.

3. The method of claim 1 or 2, wherein the attribute information includes one of a target behavior pattern, a target object, a target time, a target scene, a target topic, and a target type, the query term including one of a target behavior pattern query term, a target object query term, a target time query term, a target scene query term, a target topic query term, and a target type query term; the determining query characteristics for each of the plurality of query terms includes:

and determining target behavior mode characteristics, target object characteristics, target time characteristics, target scene characteristics, target theme characteristics and target type characteristics as query characteristics of each of the plurality of query terms.

4. The method of claim 3, wherein said determining a target file from said M initial files based on said query characteristics and said file characteristics comprises:

for each initial file, determining the correlation between the query feature and the file feature of the initial file as a cross feature; and

And determining target files from the M initial files according to the query characteristics, the file characteristics and the cross characteristics.

5. The method of claim 4, wherein the determining, for each initial file, a correlation between the query feature and a file feature of the initial file, as a cross feature, comprises: for each of the initial files,

determining the correlation between the target behavior mode characteristics and the file characteristics as behavior mode crossing characteristics;

determining the correlation between the target object feature and the file feature as an object cross feature;

determining a correlation between the target temporal feature and the file feature as a temporal intersection feature;

determining the correlation between the target scene feature and the file feature as a scene intersection feature;

determining the correlation between the target theme feature and the file feature as a theme-crossing feature; and

and determining the correlation between the target type feature and the file feature as a type crossing feature.

6. The method of claim 4 or 5, wherein said determining a target file from said M initial files based on said query features, said file features, and said cross features comprises:

According to the query term characteristics, the file characteristics and the cross characteristics, determining respective evaluation values of the M initial files;

according to the evaluation value, N candidate files are determined from the M initial files, wherein N is an integer greater than 1; and

and determining a target file from the N candidate files according to the correlation between the query text and the N candidate files.

7. The method of claim 6, wherein the determining a target file from the N candidate files based on the relevance between the query text and the N candidate files comprises:

and inputting the query text, the N candidate files and target constraint information into a large language model to obtain the target file, wherein the target constraint information is used for indicating the large language model to find out the target file most relevant to the query text from the N candidate files.

8. The method of claim 3, wherein the recalling M initial files from the plurality of query terms comprises:

recalling at least one first initial file from the database according to the plurality of query terms;

recalling at least one second initial file from a preset file set according to the plurality of query items, wherein interaction behavior information of the initial files in the preset file set is consistent with at least one of the plurality of query items;

Recalling at least one third initial file from the database according to the target subject query term; and

and determining the first initial file, the second initial file and the third initial file as the M initial files.

9. The method according to any one of claims 1 to 8, wherein extracting the sub-text belonging to each attribute from the query text according to the plurality of attribute information, and obtaining the plurality of query terms includes:

and inputting the query text, the plurality of attribute information and the query constraint information into a large language model to obtain the plurality of query items, wherein the query constraint information is used for indicating the large language model to extract the sub-texts belonging to each attribute from the query text, and determining the plurality of attribute information and the sub-texts of each attribute as the plurality of query items.

10. A document querying device, comprising:

the extraction module is used for extracting the sub-text belonging to each attribute from the query text according to the plurality of attribute information to obtain a plurality of query items;

the recall module is used for recalling M initial files according to the plurality of query items, wherein M is an integer greater than 1;

The file feature determining module is used for determining file features according to interaction behavior information associated with each initial file; and

and the target file determining module is used for determining the query characteristics of each of the plurality of query terms and determining target files from the M initial files according to the query characteristics and the file characteristics.

11. The apparatus of claim 10, wherein the interaction behavior information comprises at least one of an interaction behavior pattern, an interaction object, an interaction time, an interaction scenario, an interaction topic, and a file type; the file feature determining module is configured to determine at least one of an interaction behavior mode feature, an interaction object feature, an interaction time feature, an interaction scene feature, an interaction theme feature, and a file type feature as the file feature according to the interaction behavior information.

12. The apparatus of claim 10 or 11, wherein the attribute information comprises one of a target behavior pattern, a target object, a target time, a target scene, a target topic, and a target type, the query term comprising one of a target behavior pattern query term, a target object query term, a target time query term, a target scene query term, a target topic query term, and a target type query term; the target file determining module is further configured to determine a target behavior mode feature, a target object feature, a target time feature, a target scene feature, a target theme feature, and a target type feature, as query features of each of the plurality of query terms.

13. The apparatus of claim 12, wherein the object file determination module comprises:

a cross feature determining unit configured to determine, for each initial file, a correlation between the query feature and a file feature of the initial file as a cross feature; and

and the target file determining unit is used for determining target files from the M initial files according to the query characteristics, the file characteristics and the cross characteristics.

14. The apparatus of claim 13, wherein the cross-feature determination unit comprises:

a behavior pattern cross feature determining subunit, configured to determine a correlation between the target behavior pattern feature and the file feature as a behavior pattern cross feature;

an object cross feature determining subunit, configured to determine a correlation between the target object feature and the file feature as an object cross feature;

a time-crossing feature determination subunit configured to determine a correlation between the target time feature and the file feature as a time-crossing feature;

a scene intersection feature determining subunit, configured to determine a correlation between the target scene feature and the file feature as a scene intersection feature;

A theme-cross feature determination subunit, configured to determine a correlation between the target theme feature and the file feature as a theme-cross feature; and

and the type cross feature determining subunit is used for determining the correlation between the target type feature and the file feature as a type cross feature.

15. The apparatus according to claim 13 or 14, wherein the object file determining unit includes:

the evaluation value determining subunit is used for determining respective evaluation values of the M initial files according to the query term characteristics, the file characteristics and the cross characteristics;

a candidate file determining subunit, configured to determine N candidate files from the M initial files according to the evaluation values, where N is an integer greater than 1; and

and the target file determining subunit is used for determining the target file from the N candidate files according to the correlation between the query text and the N candidate files.

16. The apparatus of claim 15, wherein the target file determining subunit is further configured to input the query text, the N candidate files, and target constraint information into a large language model, to obtain the target file, where the target constraint information is configured to instruct the large language model to find a target file that is most relevant to the query text from the N candidate files.

17. The apparatus of claim 12, wherein the recall module comprises:

a first recall unit for recalling at least one first initial file from the database according to the plurality of query terms;

a second recall unit, configured to recall at least one second initial file from a preset file set according to the plurality of query items, where interaction behavior information of the initial files in the preset file set is consistent with at least one of the plurality of query items;

a third recall unit, configured to recall at least one third initial file from the database according to the target subject query term; and

an initial file determining unit configured to determine the first initial file, the second initial file, and the third initial file as the M initial files.

18. The apparatus according to any one of claims 10 to 17, wherein the extracting module is configured to input the query text, the plurality of attribute information, and query constraint information into a large language model, so as to obtain the plurality of query terms, where the query constraint information is configured to instruct the large language model to extract sub-texts belonging to respective attributes from the query text, and determine the plurality of attribute information and the sub-texts of the respective attributes as the plurality of query terms.

19. An electronic device, comprising:

at least one processor; and

a memory communicatively coupled to the at least one processor; wherein, the liquid crystal display device comprises a liquid crystal display device,

the memory stores instructions executable by the at least one processor to enable the at least one processor to perform the method of any one of claims 1 to 9.

20. A non-transitory computer readable storage medium storing computer instructions for causing the computer to perform the method of any one of claims 1 to 9.

21. A computer program product comprising a computer program stored on at least one of a readable storage medium and an electronic device, which, when executed by a processor, implements the method according to any one of claims 1 to 9.