CN114547286A - Information searching method and device and electronic equipment - Google Patents
Information searching method and device and electronic equipment Download PDFInfo
- Publication number
- CN114547286A CN114547286A CN202210157137.2A CN202210157137A CN114547286A CN 114547286 A CN114547286 A CN 114547286A CN 202210157137 A CN202210157137 A CN 202210157137A CN 114547286 A CN114547286 A CN 114547286A
- Authority
- CN
- China
- Prior art keywords
- ranking
- model
- sorting
- feature
- information
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
- G06F16/33—Querying
- G06F16/338—Presentation of query results
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
- G06F16/33—Querying
- G06F16/332—Query formulation
- G06F16/3329—Natural language query formulation or dialogue systems
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
- G06F16/33—Querying
- G06F16/335—Filtering based on additional data, e.g. user or group profiles
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Databases & Information Systems (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Mathematical Physics (AREA)
- Artificial Intelligence (AREA)
- Human Computer Interaction (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The application discloses an information searching method and device and electronic equipment. The method comprises the following steps: acquiring input information and candidate search results corresponding to the input information; inputting the candidate search results into N cascaded sorting models, and sorting the respective input first search results by the N cascaded sorting models respectively; aiming at any sorting model, screening a first search result of any sorting model according to a sorting result of any sorting model to obtain a second search result of any sorting model; and determining a second search result of the last ranking model in the N cascaded ranking models as a target search result corresponding to the input information. Therefore, the number of the first search results input by the sequencing model can be reduced step by step, so that the operation time of the N cascaded sequencing models is reduced step by step, the efficiency and the reliability in the information search process are improved, and the user experience is improved.
Description
Technical Field
The present application relates to the field of artificial intelligence technologies, and in particular, to an information search method and apparatus, and an electronic device.
Background
In recent years, with the development of artificial intelligence technology, ranking models have been widely applied in various application scenarios. In particular, more and more attention is paid to the manual customer service conversation scene. However, as the ranking model has more and more features to rely on in the ranking process, the time consumption of multi-dimensional feature extraction for more and more candidate sets is longer and longer, thereby significantly reducing the ranking efficiency and affecting the user experience.
In the related art, a mode of optimizing a ranking model or extracting a single-dimensional feature for optimization is usually adopted, so that the ranking efficiency is improved, and the purpose of improving the information search efficiency is further achieved. However, none of the foregoing optimization methods can effectively improve the information search efficiency.
Therefore, how to improve the response efficiency in the information search process and the accuracy and reliability of the obtained search results has become one of important research directions.
Disclosure of Invention
The present application is directed to solving, at least to some extent, one of the technical problems in the related art.
Therefore, a first objective of the present application is to provide an information searching method, which is used for solving the technical problems of low efficiency, low accuracy and poor reliability in the related information searching method process.
A second object of the present invention is to provide another information search apparatus.
A third object of the invention is to propose an electronic device.
A fourth object of the invention is to propose a computer-readable storage medium.
In order to achieve the above object, an embodiment of a first aspect of the present application provides an information searching method, where the method includes the following steps: acquiring input information and candidate search results corresponding to the input information; inputting the candidate search results into N cascaded ranking models, and ranking the respective input first search results by the N cascaded ranking models, wherein the first search results of each ranking model are second search results obtained after the previous ranking model is screened, and N is an integer greater than or equal to 1, starting from a second ranking model; aiming at any sorting model, screening the first search result of the any sorting model according to the sorting result of the any sorting model to obtain the second search result of the any sorting model; and determining the second search result of the last ranking model in the N cascaded ranking models as a target search result corresponding to the input information.
In addition, the information search method according to the above embodiment of the present application may further have the following additional technical features:
according to an embodiment of the present application, the ranking, by the N cascaded ranking models, the first search results that are respectively input includes: for any sequencing model, performing feature extraction on a first search result of any sequencing model according to a feature dimension set corresponding to any sequencing model to obtain first feature information of the first search result, wherein the feature dimension sets corresponding to the sequencing models in the N cascaded sequencing models are different; and ranking the first search result of any ranking model based on the extracted first feature information.
According to an embodiment of the present application, further comprising: the feature dimension set corresponding to the current ranking model comprises the feature dimension set corresponding to the last ranking model.
According to an embodiment of the present application, further comprising: obtaining candidate feature dimensions and feature extraction duration corresponding to the candidate feature dimensions; and dividing the candidate feature dimensions according to the feature extraction duration to generate feature dimension groups, wherein each of the N cascaded ranking models corresponds to one of the feature dimension groups.
According to an embodiment of the present application, further comprising: the feature dimension set corresponding to the current ranking model includes the feature dimension set corresponding to the previous ranking model and the feature dimension group corresponding to the current ranking model.
According to an embodiment of the present application, the dividing the candidate feature dimensions according to the feature extraction duration to generate a feature dimension group includes: sorting the feature extraction durations corresponding to all the candidate feature dimensions to obtain a sorting result; and dividing the candidate feature dimensions according to the sorting result to generate the feature dimension group.
According to an embodiment of the present application, the screening the first search result of any one of the ranking models according to the ranking result of any one of the ranking models to obtain the second search result of any one of the ranking models includes: obtaining the screening strength corresponding to any sorting model; and screening the second search result from the first search result according to the screening strength.
According to an embodiment of the present application, the obtaining the screening strength corresponding to any one of the ranking models includes: obtaining the ranking of any ranking model in the N cascaded ranking models; and obtaining the screening strength of any sequencing model according to the ranking, wherein the screening strength is positively correlated with the ranking height.
The embodiment of the first aspect of the present application provides an information search method, where input information and candidate search results corresponding to the input information are obtained, the candidate search results are input into N cascaded ranking models, the N cascaded ranking models rank respective input first search results, then, for any ranking model, the first search result of any ranking model is screened according to the ranking result of any ranking model to obtain a second search result of any ranking model, and then, the second search result of a last ranking model in the N cascaded ranking models is determined as a target search result corresponding to the input information. Therefore, the number of the first search results input by the sequencing model can be reduced step by step, so that the operation time of the N cascaded sequencing models is reduced step by step, the efficiency and the reliability in the information search process are improved, and the user experience is improved.
In order to achieve the above object, an embodiment of a second aspect of the present application provides an information searching apparatus, including: the device comprises an acquisition module, a search module and a search module, wherein the acquisition module is used for acquiring input information and candidate search results corresponding to the input information; the sorting module is used for inputting the candidate search results into N cascaded sorting models and sorting the respective input first search results by the N cascaded sorting models, wherein the first search results of each sorting model are second search results obtained after the last sorting model is screened from the second sorting model, and N is an integer greater than or equal to 1; the screening module is used for screening the first search result of any sorting model according to the sorting result of any sorting model so as to obtain the second search result of any sorting model; a determining module, configured to determine the second search result of a last ranking model in the N cascaded ranking models as a target search result corresponding to the input information.
According to an embodiment of the present application, the sorting module is further configured to: for any sequencing model, performing feature extraction on a first search result of any sequencing model according to a feature dimension set corresponding to any sequencing model to obtain first feature information of the first search result, wherein the feature dimension sets corresponding to the sequencing models in the N cascaded sequencing models are different; and ranking the first search result of any ranking model based on the extracted first feature information.
According to an embodiment of the present application, further comprising: the feature dimension set corresponding to the current ranking model comprises the feature dimension set corresponding to the last ranking model.
According to an embodiment of the present application, the sorting module is further configured to: acquiring candidate feature dimensions and feature extraction duration corresponding to the candidate feature dimensions; and dividing the candidate feature dimensions according to the feature extraction duration to generate feature dimension groups, wherein each of the N cascaded ranking models corresponds to one of the feature dimension groups.
According to an embodiment of the present application, the sorting module is further configured to: the feature dimension set corresponding to the current ranking model includes the feature dimension set corresponding to the previous ranking model and the feature dimension group corresponding to the current ranking model.
According to an embodiment of the present application, the sorting module is further configured to: sorting the feature extraction durations corresponding to all the candidate feature dimensions to obtain a sorting result; and dividing the candidate feature dimensions according to the sorting result to generate the feature dimension group.
According to an embodiment of the present application, the screening module is further configured to: obtaining the screening strength corresponding to any sorting model; and screening the second search result from the first search result according to the screening strength.
According to an embodiment of the present application, the screening module is further configured to: obtaining the ranking of any ranking model in the N cascaded ranking models; and obtaining the screening strength of any sequencing model according to the ranking, wherein the screening strength is positively correlated with the ranking height.
The second search result of any ranking model is obtained by screening the first search result of any ranking model according to the ranking result of any ranking model aiming at any ranking model and according to the ranking result of any ranking model, and then the second search result of the last ranking model in the N cascaded ranking models is determined as the target search result corresponding to the input information. Therefore, the number of the first search results input by the sequencing model can be reduced step by step, so that the operation time of the N cascaded sequencing models is reduced step by step, the efficiency and the reliability in the information search process are improved, and the user experience is improved.
In order to achieve the above object, an embodiment of a third aspect of the present application provides an electronic device, including: a memory, a processor and a computer program stored on the memory and operable on the processor, the processor implementing the information search method as described in any of the embodiments of the first aspect of the present application when executing the program.
In order to achieve the above object, a fourth aspect of the present application provides a computer-readable storage medium, where the program is executed by a processor to implement the information search method as described in any one of the embodiments of the first aspect of the present application.
Drawings
Fig. 1 is a schematic flowchart of an information search method according to an embodiment of the present application;
FIG. 2 is a schematic flow chart diagram illustrating an information search method according to another embodiment of the present disclosure;
FIG. 3 is a schematic flow chart diagram illustrating an information search method according to another embodiment of the present disclosure;
FIG. 4 is a schematic flow chart diagram illustrating an information search method according to another embodiment of the present disclosure;
FIG. 5 is a schematic flow chart diagram illustrating an information search method according to another embodiment of the present disclosure;
FIG. 6 is a schematic flow chart diagram illustrating an information search method according to another embodiment of the present disclosure;
FIG. 7 is a schematic flow chart diagram illustrating an information search method according to another embodiment of the present disclosure;
FIG. 8 is a schematic structural diagram of an information search apparatus according to an embodiment of the present application;
fig. 9 is a schematic structural diagram of an electronic device according to an embodiment of the present application.
Detailed Description
For a better understanding of the above technical solutions, exemplary embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be embodied in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the disclosure to those skilled in the art.
It should be understood that "and/or" referred to in the embodiments of the present application describes an association relationship of associated objects, and indicates that three relationships may exist, for example, a and/or B may indicate: a exists alone, A and B exist simultaneously, and B exists alone, wherein A and B can be singular or plural. The character "/" generally indicates that the former and latter associated objects are in an "or" relationship.
An information search method, an information search device, and an electronic device according to embodiments of the present application are described below with reference to the accompanying drawings.
Fig. 1 is a schematic flowchart of an information search method according to an embodiment of the present application.
As shown in fig. 1, the duration prediction method provided in the embodiment of the present application specifically includes the following steps:
s101, obtaining input information and candidate search results corresponding to the input information.
The input information may be any question input by the user. For example, the input information may be "is the refrigerator consume high? ".
The candidate search results corresponding to the input information may be all search results obtained by querying according to the input information.
It should be noted that the number of candidate search results may be identified after the candidate search results are obtained, and the candidate search results are obtained again after it is determined that the number of candidate search results does not reach the preset number.
For example, if the preset number is set to 5, in this case, if the input information "is that the refrigerator consumes more power? "there are 4 candidate search results, the candidate search results can be retrieved according to the input information.
It should be noted that, in the present application, a specific manner of obtaining the candidate search result according to the input information query is not limited, and may be selected according to an actual situation.
For example, a target keyword may be obtained from the input information, and the target keyword is obtained according to the similarity between the target keyword and a keyword in a question in a preset database, and an answer corresponding to a question with the similarity reaching a similarity threshold is selected as a candidate search result.
S102, inputting the candidate search results into N cascaded ranking models, and ranking the respective input first search results by the N cascaded ranking models, wherein the first search results of each ranking model are second search results obtained after the last ranking model is screened from the second ranking model, and N is an integer greater than or equal to 1.
In the present application, the structure of the N cascaded ranking models is not limited, and may be set according to actual situations. For example, N cascaded ranking models may be set as 2 structurally identical ranking models, and the 2 ranking models are connected in a Cascade (Cascade) manner.
Further, the candidate search results may be input into N cascaded ranking models, and the N cascaded ranking models rank the respective input first search results.
Wherein, for the first ranking model, the first search result may be a candidate search result; and starting from the second sequencing model, the first search result of each sequencing model is the second search result after the last sequencing model is screened.
For example, for a total of 10 candidate search results a to J, the 10 candidate search results are input into a first ranking model for ranking to obtain a second search result output by the first ranking model. And further, inputting the second search result output by the first sorting model into a second sorting model for sorting to obtain a second search result output by the second sorting model.
S103, aiming at any sorting model, screening the first search result of any sorting model according to the sorting result of any sorting model to obtain a second search result of any sorting model.
In the embodiment of the application, for any ranking model, the first search result of any ranking model can be screened according to the ranking result of any ranking model, so as to obtain the second search result of any ranking model.
It should be noted that, in this case, the number of output second search results is smaller than the number of input first search results of the model, which significantly reduces the operation time consumption of the next-level ranking model.
In the present application, the specific manner of filtering the first search result is not limited, and may be set according to actual situations.
Optionally, after the first search result of any ranking model is obtained, the first search result within the preset range may be selected as the second search result, and the first search result not within the preset range is discarded.
For example, for a-J, 10 candidate search results are input into the first ranking model for ranking, the ranking result (positive ranking) is A, B, C, H, I, J, K, D, E, F, in which case, if the predetermined range is TOP 4 (TOP 4 bits), the second search result of the ranking model is A, B, C, H.
And S104, determining a second search result of the last ranking model in the N cascaded ranking models as a target search result corresponding to the input information.
For example, for 2 cascaded ranking models, the second search result of the second ranking model may be determined as the target search result corresponding to the input information.
Therefore, the information searching method provided by the application can obtain the input information and the candidate searching results corresponding to the input information, input the candidate searching results into the N cascaded sorting models, sort the respective input first searching results by the N cascaded sorting models, and then screen the first searching results of any sorting model according to the sorting results of any sorting model aiming at any sorting model to obtain the second searching results of any sorting model, thereby determining the second searching results of the last sorting model in the N cascaded sorting models as the target searching results corresponding to the input information. Therefore, the number of the first search results input by the sequencing model can be reduced step by step, so that the operation time of the N cascaded sequencing models is reduced step by step, the efficiency and the reliability in the information search process are improved, and the user experience is improved.
In the present application, when trying to sort the first search results respectively input by the N cascaded ranking models, feature extraction may be performed on the first search results, and then the extracted feature information may be ranked.
As a possible implementation manner, as shown in fig. 2, on the basis of the foregoing embodiment, a specific process of sorting the first search results respectively input by N cascaded sorting models in the foregoing step includes the following steps:
s201, for any sequencing model, performing feature extraction on the first search result of any sequencing model according to a feature dimension set corresponding to any sequencing model to obtain first feature information of the first search result, wherein feature dimension sets corresponding to the sequencing models in the N cascaded sequencing models are different.
In the present application, when attempting to perform feature extraction, feature dimension sets corresponding to respective ranking models are different.
Optionally, the feature dimension set corresponding to the current ranking model includes a feature dimension set corresponding to a previous ranking model; optionally, the feature dimension set corresponding to the current ranking model includes a part of feature dimensions corresponding to the last ranking model.
For example, for the current ranking model (the second ranking model), the corresponding feature dimension set is T1+ T2, and the feature dimension set T1 corresponding to the last ranking model (the first ranking model) is included; for the current ranking model (the second ranking model), the corresponding feature dimension set is T1 '+ T2, and the partial feature dimension T1' in the feature dimension set T1 corresponding to the previous ranking model (the first ranking model) is included.
The feature dimension set may include any of the following feature dimensions: word-level JaccardWord (jaccarded similarity) dimension, Word-level JaccardW2v (jaccarded similarity) dimension in combination with computation of Word vectors, binary or ternary Word-level jaccardgram (jaccarded similarity) dimension, W2v (Word to Vector) dimension in which similarity is computed using Word vectors, Longest Common Subsequence (LCS) dimension, and feature dimension in which similarity is computed based on patterns such as bert (binary Encoder responses from transformations), esim (enhanced Sequential information model), dssm (deep Structured Semantic models), cdssm (relational language model), etc.
It should be noted that, in the present application, a specific selection manner of the feature dimension set is not limited, and may be selected according to an actual situation.
Alternatively, the candidate feature dimensions may be directly partitioned to generate the set of feature dimensions.
For example, the candidate feature dimensions 1-10 can be divided into the following 4 feature dimension groups, which are respectively: a first characteristic dimension group comprising characteristic dimensions 1-3; a second set of feature dimensions comprising feature dimensions 4-6; a third set of feature dimensions comprising feature dimensions 7-8, and a fourth set of feature dimensions comprising feature dimensions 9-10. In this case, the feature dimension set corresponding to the first ranking model may be determined as a first feature dimension group; the feature dimension set corresponding to the second ranking model is a first feature dimension group + a second feature dimension group; the feature dimension set corresponding to the third ranking model is a first feature dimension group, a second feature dimension group and a third feature dimension group; the feature dimension set corresponding to the fourth ranking model is the first feature dimension group + the second feature dimension group + the third feature dimension group + the fourth feature dimension group.
Optionally, the candidate feature dimensions may be divided according to the feature extraction durations corresponding to the candidate feature dimensions to generate the feature dimension group. The feature dimension set corresponding to the current sorting model comprises a feature dimension set corresponding to a last sorting model and a feature dimension group corresponding to the current sorting model.
As a possible implementation manner, as shown in fig. 3, on the basis of the foregoing embodiment, the method specifically includes the following steps:
s301, obtaining candidate feature dimensions and feature extraction duration corresponding to the candidate feature dimensions.
The feature extraction duration is one of statistical information corresponding to the candidate feature dimensions, and the feature extraction duration can be obtained by inquiring the statistical information.
For example, for candidate feature dimensions 1-10, the corresponding feature extraction durations are 10s, 13s, 11s, 15s, 12s, 20s, 7s, and 8 s.
S302, dividing the candidate feature dimensions according to the feature extraction duration to generate feature dimension groups, wherein each of the N cascaded ranking models corresponds to one feature dimension group.
In the embodiment of the application, after the corresponding feature extraction duration is obtained, the candidate feature dimensions can be sorted according to the feature extraction duration. Further, the candidate feature dimensions may be divided according to the ranking result to generate a feature dimension group.
As a possible implementation manner, as shown in fig. 4, on the basis of the foregoing embodiment, a specific process of dividing candidate feature dimensions according to a feature extraction duration in the foregoing step to generate a feature dimension group includes the following steps:
s401, sorting the feature extraction durations corresponding to all the candidate feature dimensions to obtain a sorting result.
S402, dividing the candidate feature dimensions according to the sorting result to generate a feature dimension group.
Optionally, the candidate feature dimension that consumes the shortest time may be used as the feature dimension set of the top ranking model, and the candidate feature dimension that consumes the longest time may be used as the feature dimension set of the unpicked ranking model.
For example, for candidate feature dimensions 1-10, the corresponding feature extraction durations are 10s, 13s, 11s, 15s, 12s, 20s, 7s, and 8 s. Further, the candidate feature dimensions may be sorted (e.g., in positive order) according to the corresponding feature extraction duration, and the candidate feature dimensions 9, 10, 1, 2, 4, 7, 3, 5, 6, and 8. In this case, all candidate feature dimensions may be divided into the following 4 feature dimension groups, which are: a first set of feature dimensions comprising feature dimensions 9, 10, 1; a second set of feature dimensions comprising feature dimensions 2, 4, 7; a third set of feature dimensions comprising feature dimensions 3, 5, and a fourth set of feature dimensions comprising feature dimensions 6, 8. In this case, the feature dimension set corresponding to the first ranking model may be determined as a first feature dimension group; the feature dimension set corresponding to the second ranking model is a first feature dimension group, the second feature dimension group and the third feature dimension group; the feature dimension set corresponding to the fourth ranking model is the first feature dimension group + the second feature dimension group + the third feature dimension group + the fourth feature dimension group.
S202, ranking the first search results of any ranking model based on the extracted first feature information.
Therefore, according to the information search method provided by the application, the first search result of any sorting model can be subjected to feature extraction according to the feature dimension set corresponding to any sorting model aiming at any sorting model so as to obtain the first feature information of the first search result, and then the first search result of any sorting model is sorted based on the extracted first feature information. Therefore, the candidate feature dimensions can be reasonably divided based on the statistical information such as the time length of feature extraction, so that the feature dimension set corresponding to any one sequencing model is obtained, after the previous sequencing model is roughly arranged, sequencing can be performed more accurately by the next sequencing model, the efficiency and the reliability in the information searching process are further improved, and the user experience is improved.
It should be noted that, in the present application, when trying to obtain the second search result according to the sorting result, the second search result may be filtered from the first search result according to a preset filtering strength.
As a possible implementation manner, as shown in fig. 5, on the basis of the foregoing embodiment, a specific process of filtering the first search result of any ranking model according to the ranking result of any ranking model in the foregoing step to obtain the second search result of any ranking model includes the following steps:
s501, obtaining the screening strength corresponding to any sorting model.
In the present application, the setting manner of the screening strength corresponding to each ranking model is not limited, and may be selected according to actual situations.
Optionally, a unique filtering strength may be set for the N ranking models.
Alternatively, different filtering strengths may be set for the N ranking models.
As a possible implementation manner, as shown in fig. 6, on the basis of the foregoing embodiment, a specific process of obtaining a screening strength corresponding to any one ranking model in the foregoing steps includes the following steps:
s601, obtaining the ranking of any ranking model in the N cascaded ranking models.
S602, obtaining the screening strength of any sequencing model according to the ranking, wherein the screening strength is positively correlated with the ranking.
For example, for 2 cascaded ranking models, the screening strength of the first ranking model is 60% and the screening strength of the second ranking model is 50%.
S502, screening a second search result from the first search result according to the screening strength.
For example, for a first ranking model of 2 cascaded ranking models, if the screening strength is 60% and the number of first search results is 10, the number of second search results is 6; for the second ranking model, the screening strength is 50%, the first search results are 6, and the second search results are 3.
Therefore, the information searching method provided by the application can be used for screening the second searching result from the first searching result according to the screening strength by obtaining the screening strength corresponding to any sorting model. Therefore, the method and the device can be set based on reasonable screening strength, time consumption in the information searching process is further shortened, efficiency and reliability in the information searching process are further improved, and user experience is improved.
Fig. 7 is a flowchart illustrating an information search method according to another embodiment of the present application.
As shown in fig. 7, a flowchart of the information search method provided in the embodiment of the present application specifically includes the following steps:
s701, acquiring input information and candidate search results corresponding to the input information.
S702, obtaining candidate feature dimensions and feature extraction duration corresponding to the candidate feature dimensions.
And S703, sorting the feature extraction durations corresponding to all the candidate feature dimensions to obtain a sorting result.
And S704, dividing the candidate feature dimensions according to the sorting result to generate a feature dimension group.
S705, acquiring a feature dimension set corresponding to any sequencing model according to the feature dimension group.
S706, for any one of the ranking models, according to the feature dimension set corresponding to any one of the ranking models, feature extraction is performed on the first search result of any one of the ranking models to obtain first feature information of the first search result.
And S707, ranking the first search results of any ranking model based on the extracted first characteristic information.
S708, obtaining the ranking of any ranking model in the N cascaded ranking models.
S709, obtaining the screening strength of any sequencing model according to the ranking, wherein the screening strength is positively correlated with the ranking.
And S710, aiming at any sorting model, screening a second search result from the first search result according to the screening strength.
And S711, determining a second search result of the last ranking model in the N cascaded ranking models as a target search result corresponding to the input information.
It should be noted that the information search method provided by the present application is applicable to various application scenarios.
Taking a manual customer service conversation scene as an example, the input information input by the user is "is the refrigerator consumes high power? "obtain 100 candidate search results in total of 1-100, input 100 candidate search results into 2 cascaded ranking models, the first ranking model may perform feature extraction on 100 candidate search results through a feature extraction network to obtain 100 corresponding first feature information, then rank and screen 100 first feature information through a ranking network in the first ranking model, the screening strength is 40%, in this case, for the first ranking model, 40 second search results in total are obtained.
Further, 40 second search results are input into a second ranking model, feature extraction is performed on the 40 candidate search results by a feature extraction network in the second ranking model to obtain corresponding 40 first feature information, then the 40 first feature information is ranked and screened by the ranking network in the second ranking model, the screening strength is 10%, and under the condition, 4 second search results are obtained for the second ranking model.
Further, since the second ranking model is the last ranking model, 4 second search results of the second ranking model may be determined as target search results corresponding to the input information.
Therefore, the information searching method provided by the application can reduce the number of the first searching results input by the sequencing model step by step, so that the operation time consumption of the N cascaded sequencing models is reduced step by step, the efficiency and the reliability in the information searching process are improved, and the user experience is improved. Furthermore, based on statistical information such as the time length of feature extraction, candidate feature dimensions are reasonably divided, so that a feature dimension set corresponding to any one sequencing model is obtained, after the previous sequencing model is roughly arranged, sequencing can be carried out more accurately by the next sequencing model, based on reasonable screening force setting, time consumption in the information searching process is further shortened, the efficiency and the reliability in the information searching process are further improved, and user experience is improved.
Based on the same application concept, the embodiment of the application also provides a device corresponding to the information searching method.
Fig. 8 is a schematic structural diagram of an information search apparatus according to an embodiment of the present application.
As shown in fig. 8, the information search apparatus 1000 includes: an acquisition module 110, a ranking module 120, a screening module 130, and a determination module 140. Wherein the content of the first and second substances,
an obtaining module 110, configured to obtain input information and a candidate search result corresponding to the input information;
a sorting module 120, configured to input the candidate search results into N cascaded sorting models, and sort the input first search results by the N cascaded sorting models, where, starting from a second sorting model, the first search result of each sorting model is a second search result obtained after the last sorting model is screened, and N is an integer greater than or equal to 1;
a screening module 130, configured to, for any ranking model, screen the first search result of the any ranking model according to the ranking result of the any ranking model, so as to obtain the second search result of the any ranking model;
a determining module 140, configured to determine the second search result of the last ranking model in the N cascaded ranking models as a target search result corresponding to the input information.
According to an embodiment of the present application, the sorting module 120 is further configured to:
for any sequencing model, performing feature extraction on a first search result of any sequencing model according to a feature dimension set corresponding to any sequencing model to obtain first feature information of the first search result, wherein the feature dimension sets corresponding to the sequencing models in the N cascaded sequencing models are different;
and ranking the first search result of any ranking model based on the extracted first feature information.
According to the embodiment of the application, the method further comprises the following steps: the feature dimension set corresponding to the current ranking model comprises the feature dimension set corresponding to the last ranking model.
According to an embodiment of the present application, the sorting module 120 is further configured to:
obtaining candidate feature dimensions and feature extraction duration corresponding to the candidate feature dimensions;
and dividing the candidate feature dimensions according to the feature extraction duration to generate feature dimension groups, wherein each of the N cascaded ranking models corresponds to one of the feature dimension groups.
According to an embodiment of the present application, the sorting module 120 is further configured to:
the feature dimension set corresponding to the current ranking model includes the feature dimension set corresponding to the previous ranking model and the feature dimension group corresponding to the current ranking model.
According to an embodiment of the present application, the sorting module 120 is further configured to:
sorting the feature extraction durations corresponding to all the candidate feature dimensions to obtain a sorting result;
and dividing the candidate feature dimensions according to the sorting result to generate the feature dimension group.
According to an embodiment of the present application, the screening module 130 is further configured to:
obtaining the screening strength corresponding to any sorting model;
and screening the second search result from the first search result according to the screening strength.
According to an embodiment of the present application, the screening module 130 is further configured to:
obtaining the ranking of any ranking model in the N cascaded ranking models;
and obtaining the screening strength of any sequencing model according to the ranking, wherein the screening strength is positively correlated with the ranking height.
Therefore, the information searching device provided by the application can obtain the input information and the candidate searching results corresponding to the input information, input the candidate searching results into the N cascaded sorting models, sort the respective input first searching results by the N cascaded sorting models, and then screen the first searching results of any sorting model according to the sorting results of any sorting model aiming at any sorting model to obtain the second searching results of any sorting model, thereby determining the second searching results of the last sorting model in the N cascaded sorting models as the target searching results corresponding to the input information. Therefore, the number of the first search results input by the sequencing model can be reduced step by step, so that the operation time of the N cascaded sequencing models is reduced step by step, the efficiency and the reliability in the information search process are improved, and the user experience is improved.
Based on the same application concept, the embodiment of the application also provides the electronic equipment.
Fig. 9 is a schematic structural diagram of an electronic device according to an embodiment of the present application. As shown in fig. 9, the electronic device 3000 includes a memory 310, a processor 320, and a computer program stored in the memory 310 and operable on the processor 320, and when the processor executes the computer program, the information search method is implemented.
As will be appreciated by one skilled in the art, embodiments of the present application may be provided as a method, system, or computer program product. Accordingly, the present application may take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, the present application may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, and the like) having computer-usable program code embodied therein.
The present application is described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the application. It will be understood that each flow and/or block of the flowchart illustrations and/or block diagrams, and combinations of flows and/or blocks in the flowchart illustrations and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means which implement the function specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
It should be noted that in the claims, any reference signs placed between parentheses shall not be construed as limiting the claim. The word "comprising" does not exclude the presence of elements or steps not listed in a claim. The word "a" or "an" preceding an element does not exclude the presence of a plurality of such elements. The application can be implemented by means of hardware comprising several distinct elements, and by means of a suitably programmed computer. In the unit claims enumerating several means, several of these means may be embodied by one and the same item of hardware. The usage of the words first, second and third, etcetera do not indicate any ordering. These words may be interpreted as names.
While the preferred embodiments of the present application have been described, additional variations and modifications in those embodiments may occur to those skilled in the art once they learn of the basic inventive concepts. Therefore, it is intended that the appended claims be interpreted as including preferred embodiments and all alterations and modifications as fall within the scope of the application.
It will be apparent to those skilled in the art that various changes and modifications may be made in the present invention without departing from the spirit and scope of the invention. Thus, if such modifications and variations of the present application fall within the scope of the claims of the present application and their equivalents, the present application is intended to include such modifications and variations as well.
Claims (18)
1. An information search method, comprising:
acquiring input information and candidate search results corresponding to the input information;
inputting the candidate search results into N cascaded ranking models, and ranking the respective input first search results by the N cascaded ranking models, wherein the first search results of each ranking model are second search results obtained after the previous ranking model is screened, and N is an integer greater than or equal to 1, starting from a second ranking model;
aiming at any sequencing model, screening the first search result of any sequencing model according to the sequencing result of any sequencing model to obtain the second search result of any sequencing model;
and determining the second search result of the last ranking model in the N cascaded ranking models as a target search result corresponding to the input information.
2. The information search method according to claim 1, wherein the sorting the respective input first search results by the N cascaded sorting models respectively comprises:
for any sequencing model, performing feature extraction on a first search result of any sequencing model according to a feature dimension set corresponding to any sequencing model to obtain first feature information of the first search result, wherein the feature dimension sets corresponding to the sequencing models in the N cascaded sequencing models are different;
and ranking the first search result of any ranking model based on the extracted first feature information.
3. The information search method according to claim 2, further comprising:
the feature dimension set corresponding to the current ranking model comprises the feature dimension set corresponding to the last ranking model.
4. The information search method according to claim 3, further comprising:
obtaining candidate feature dimensions and feature extraction duration corresponding to the candidate feature dimensions;
and dividing the candidate feature dimensions according to the feature extraction duration to generate feature dimension groups, wherein each of the N cascaded ranking models corresponds to one of the feature dimension groups.
5. The information search method according to claim 4, further comprising:
the feature dimension set corresponding to the current ranking model includes the feature dimension set corresponding to the previous ranking model and the feature dimension group corresponding to the current ranking model.
6. The information search method according to claim 4, wherein the dividing the candidate feature dimensions according to the feature extraction duration to generate a feature dimension group comprises:
sorting the feature extraction durations corresponding to all the candidate feature dimensions to obtain a sorting result;
and dividing the candidate feature dimensions according to the sorting result to generate the feature dimension group.
7. The information search method according to claim 1, wherein the filtering the first search result of any ranking model according to the ranking result of any ranking model to obtain the second search result of any ranking model comprises:
obtaining the screening strength corresponding to any sorting model;
and screening the second search result from the first search result according to the screening strength.
8. The information search method according to claim 7, wherein the obtaining of the screening strength corresponding to any one of the ranking models includes:
obtaining the ranking of any ranking model in the N cascaded ranking models;
and obtaining the screening strength of any sequencing model according to the ranking, wherein the screening strength is positively correlated with the ranking height.
9. An information search apparatus, comprising:
the device comprises an acquisition module, a search module and a search module, wherein the acquisition module is used for acquiring input information and candidate search results corresponding to the input information;
the sorting module is used for inputting the candidate search results into N cascaded sorting models, and sorting the respective input first search results by the N cascaded sorting models respectively, wherein the first search results of each sorting model are second search results obtained after the last sorting model is screened from the second sorting model, and N is an integer greater than or equal to 1;
the screening module is used for screening the first search result of any sorting model according to the sorting result of any sorting model so as to obtain the second search result of any sorting model;
a determining module, configured to determine the second search result of a last ranking model in the N cascaded ranking models as a target search result corresponding to the input information.
10. The information searching apparatus of claim 9, wherein the ranking module is further configured to:
for any sequencing model, performing feature extraction on a first search result of any sequencing model according to a feature dimension set corresponding to any sequencing model to obtain first feature information of the first search result, wherein the feature dimension sets corresponding to the sequencing models in the N cascaded sequencing models are different;
and ranking the first search result of any ranking model based on the extracted first feature information.
11. The information search apparatus according to claim 10, further comprising:
the feature dimension set corresponding to the current ranking model comprises the feature dimension set corresponding to the last ranking model.
12. The information searching apparatus of claim 11, wherein the ranking module is further configured to:
obtaining candidate feature dimensions and feature extraction duration corresponding to the candidate feature dimensions;
and dividing the candidate feature dimensions according to the feature extraction duration to generate feature dimension groups, wherein each of the N cascaded ranking models corresponds to one of the feature dimension groups.
13. The information searching apparatus of claim 12, wherein the ranking module is further configured to:
the feature dimension set corresponding to the current ranking model includes the feature dimension set corresponding to the previous ranking model and the feature dimension group corresponding to the current ranking model.
14. The information searching apparatus of claim 12, wherein the ranking module is further configured to:
sorting the feature extraction durations corresponding to all the candidate feature dimensions to obtain a sorting result;
and dividing the candidate feature dimensions according to the sorting result to generate the feature dimension group.
15. The information search apparatus of claim 9, wherein the filtering module is further configured to:
obtaining the screening strength corresponding to any sorting model;
and screening the second search result from the first search result according to the screening strength.
16. The information search device of claim 15, wherein the filtering module is further configured to:
obtaining the ranking of any ranking model in the N cascaded ranking models;
and obtaining the screening strength of any sequencing model according to the ranking, wherein the screening strength is positively correlated with the ranking height.
17. An electronic device, comprising: memory, processor and computer program stored on the memory and executable on the processor, which when executed by the processor implements the information search method according to any one of claims 1 to 8.
18. A computer-readable storage medium, on which a computer program is stored, which, when being executed by a processor, carries out an information search method according to any one of claims 1 to 8.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210157137.2A CN114547286A (en) | 2022-02-21 | 2022-02-21 | Information searching method and device and electronic equipment |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210157137.2A CN114547286A (en) | 2022-02-21 | 2022-02-21 | Information searching method and device and electronic equipment |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN114547286A true CN114547286A (en) | 2022-05-27 |
Family
ID=81675153
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210157137.2A Pending CN114547286A (en) | 2022-02-21 | 2022-02-21 | Information searching method and device and electronic equipment |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114547286A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2024088231A1 (en) * | 2022-10-28 | 2024-05-02 | 华为技术有限公司 | Signal processing method and apparatus, and device, medium and chip |
-
2022
- 2022-02-21 CN CN202210157137.2A patent/CN114547286A/en active Pending
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2024088231A1 (en) * | 2022-10-28 | 2024-05-02 | 华为技术有限公司 | Signal processing method and apparatus, and device, medium and chip |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN112199375B (en) | Cross-modal data processing method and device, storage medium and electronic device | |
| CN108804641B (en) | Text similarity calculation method, device, equipment and storage medium | |
| CN109189991B (en) | Duplicate video identification method, device, terminal and computer readable storage medium | |
| CN106547887B (en) | Search recommendation method and device based on artificial intelligence | |
| CN111047563B (en) | Neural network construction method applied to medical ultrasonic image | |
| CN109033244B (en) | Search result ordering method and device | |
| CN112035599B (en) | Query method and device based on vertical search, computer equipment and storage medium | |
| CN110134777B (en) | Question duplication eliminating method and device, electronic equipment and computer readable storage medium | |
| CN109753577B (en) | Method and related device for searching human face | |
| CN108182200B (en) | Keyword expansion method and device based on semantic similarity | |
| CN116431837B (en) | Document retrieval method and device based on large language model and graph network model | |
| CN112699945A (en) | Data labeling method and device, storage medium and electronic device | |
| CN105989001A (en) | Image searching method and device, and image searching system | |
| CN112364014A (en) | Data query method, device, server and storage medium | |
| CN114547286A (en) | Information searching method and device and electronic equipment | |
| CN114037007A (en) | Data set construction method and device, computer equipment and storage medium | |
| CN113515620A (en) | Method and device for sorting technical standard documents of power equipment, electronic equipment and medium | |
| CN106407332B (en) | Search method and device based on artificial intelligence | |
| CN113821657A (en) | Artificial intelligence-based image processing model training method and image processing method | |
| CN108170664B (en) | Key word expansion method and device based on key words | |
| CN108830302B (en) | Image classification method, training method, classification prediction method and related device | |
| CN111191065A (en) | Homologous image determining method and device | |
| CN110442681A (en) | A kind of machine reads method, electronic equipment and the readable storage medium storing program for executing of understanding | |
| CN114048148A (en) | Crowdsourcing test report recommendation method and device and electronic equipment | |
| CN109614542B (en) | Public number recommendation method, device, computer equipment and storage medium |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |