Search

US-12619617-B2 - Information recommendation method, apparatus, electronic device, and storage medium

US12619617B2US 12619617 B2US12619617 B2US 12619617B2US-12619617-B2

Abstract

An information recommendation method, an information recommendation apparatus, an electronic device, and a storage medium are provided. The information recommendation method includes: obtaining a candidate recommendation information set which matches current search information, the candidate recommendation information set including multiple candidate recommendation information; determining, according to search volumes of the multiple candidate recommendation information within different search time limitations, sequence feature information of the candidate recommendation information set under each search time limitation, the sequence feature information being used to reflect an overall search feature corresponding to the candidate recommendation information set; and determining target recommendation information based on a trained target ranking model and the sequence feature information of the candidate recommendation information set under each search time limitation.

Inventors

  • Bofan XUE

Assignees

  • BEIJING ZITIAO NETWORK TECHNOLOGY CO., LTD.

Dates

Publication Date
20260505
Application Date
20240327
Priority Date
20230328

Claims (19)

  1. 1 . An information recommendation method, implemented by a terminal device, comprising: receiving input information in a search box of an application program as current search information through a GUI (graphical user interface) of the terminal device, and obtaining a candidate recommendation information set which matches the current search information, wherein the candidate recommendation information set comprises multiple candidate recommendation information, the multiple candidate recommendation information comes from multi-path recall, recall sources of the multi-path recall comprise at least one of selected from the group consisting of: user behavior of a user terminal matched with the current search information, social contact of the user terminal, and historical search mode, wherein data of the recall sources is stored in a storage device, and the multiple candidate recommendation information is obtained by accessing the storage device; determining, according to search volumes of the multiple candidate recommendation information within different search time limitations, sequence feature information of the candidate recommendation information set under each search time limitation, wherein the sequence feature information is used to reflect an overall search feature corresponding to the candidate recommendation information set; determining target recommendation information based on a trained target ranking model and the sequence feature information of the candidate recommendation information set under each search time limitation; and displaying the target recommendation information on the GUI; wherein the determining target recommendation information based on the trained target ranking model and the sequence feature information of the candidate recommendation information set under each search time limitation, comprises: inputting, by the terminal device, the sequence feature information of the candidate recommendation information set under each search time limitation to the trained target ranking model that is run in the terminal device, to obtain an estimated click-through rate corresponding to each candidate recommendation information in the candidate recommendation information set; based on the estimated click-through rate corresponding to each candidate recommendation information, selecting a predetermined number of candidate recommendation information as the target recommendation information; wherein the trained target ranking model comprises a first ranking model and a second ranking model, and a model accuracy of the second ranking model is greater than a model accuracy of the first ranking model, inputting, by the terminal device, the sequence feature information of the candidate recommendation information set under each search time limitation to the trained target ranking model that is run in the terminal device, to obtain the estimated click-through rate corresponding to each candidate recommendation information in the candidate recommendation information set, comprises: selecting partial feature information in a feature information set comprising the sequence feature information of the candidate recommendation information set under each search time limitation, and inputting the selected partial feature information into the first ranking model running in the terminal device, and predicting an initial click-through rate of each candidate recommendation information; determining a first ranking result of the multiple candidate recommendation information according to the initial click-through rate corresponding to each candidate recommendation information; and determining, based on the first ranking result and the sequence feature information of the candidate recommendation information set under each search time limitation, the estimated click-through rate corresponding to each candidate recommendation information by using the second ranking model running in the terminal device.
  2. 2 . The method according to claim 1 , wherein the sequence feature information comprises ranking feature information and variance feature information, the ranking feature information is used to reflect a ranking feature of the multiple candidate recommendation information in the candidate recommendation information set in terms of a size of search volume, and the variance feature information is used to reflect a deviation degree of the search volumes of the multiple candidate recommendation information with respect to an average search volume corresponding to the candidate recommendation information set.
  3. 3 . The method according to claim 2 , wherein the variance feature information is determined according to steps comprising: for each search time limitation, determining an average search volume under the search time limitation according to search volumes of the multiple candidate recommendation information in the candidate recommendation information set within the search time limitation; determining a search volume standard deviation under the search time limitation based on the average search volume and a search volume of each of the multiple candidate recommendation information within the search time limitation; and determining variance feature information of the candidate recommendation information set under the search time limitation based on the search volume standard deviation, the average search volume, and the search volumes of the multiple candidate recommendation information within the search time limitation.
  4. 4 . The method according to claim 3 , wherein determining the variance feature information of the candidate recommendation information set under the search time limitation based on the search volume standard deviation, the average search volume, and the search volumes of the multiple candidate recommendation information within the search time limitation comprises: for each candidate recommendation information, determining a search volume difference value based on the average search volume and a search volume of the candidate recommendation information within the search time limitation, and dividing the search volume difference value by the search volume standard deviation to obtain a search volume ratio value of the candidate recommendation information; and determining the variance feature information of the candidate recommendation information set under the search time limitation based on search volume ratio values to which the multiple candidate recommendation information corresponds.
  5. 5 . The method according to claim 1 , wherein determining the target recommendation information based on the trained target ranking model and the sequence feature information of the candidate recommendation information set under each search time limitation comprises: determining candidate feature information of each candidate recommendation information in the candidate recommendation information set, wherein the candidate feature information comprises text feature information and/or vertical category feature information; obtaining historical feature information authorized by a user, wherein the historical feature information comprises behavior feature information and/or background feature information; and determining, by using the trained target ranking model, the target recommendation information based on the historical feature information, the candidate feature information of each candidate recommendation information in the candidate recommendation information set, and the sequence feature information of the candidate recommendation information set under each search time limitation.
  6. 6 . The method according to claim 5 , wherein the feature information set further comprises the historical feature information, and the candidate feature information of each candidate recommendation information in the candidate recommendation information set, and determining, by using the trained target ranking model, the target recommendation information based on the historical feature information, the candidate feature information of each candidate recommendation information in the candidate recommendation information set, and the sequence feature information of the candidate recommendation information set under each search time limitation comprises: selecting the partial feature information from the feature information set; determining, based on the partial feature information and the first ranking model, the first ranking result of the multiple candidate recommendation information in the candidate recommendation information set; determining, by using the second ranking model, a second ranking result of the multiple candidate recommendation information in the candidate recommendation information set based on the first ranking result, the historical feature information, the candidate feature information of each candidate recommendation information in the candidate recommendation information set, and the sequence feature information of the candidate recommendation information set under each search time limitation; and determining, according to the second ranking result, the target recommendation information from the candidate recommendation information set.
  7. 7 . The method according to claim 6 , wherein determining, according to the second ranking result, the target recommendation information from the candidate recommendation information set comprises: selecting, according to ranked positions of the multiple candidate recommendation information as indicated by the second ranking result, a preset number of candidate recommendation information from the candidate recommendation information set, and determining the preset number of candidate recommendation information as the target recommendation information; and/or selecting, according to the estimated click-through rate of each candidate recommendation information as indicated by the second ranking result, candidate recommendation information with an estimated click-through rate greater than a threshold value from the candidate recommendation information set, and determining the candidate recommendation information with the estimated click-through rate greater than the threshold value as the target recommendation information.
  8. 8 . An electronic device, comprising: a processor; and a memory, being in communication connection to the processor, wherein one or more computer-executable instructions are stored on the memory, and the one or more computer-executable instructions, when executed by the processor, cause the processor to perform steps of an information recommendation method; wherein the information recommendation method comprises: receiving input information in a search box of an application program as current search information through a GUI (graphical user interface) of the electronic device, and obtaining a candidate recommendation information set which matches the current search information, wherein the candidate recommendation information set comprises multiple candidate recommendation information, the multiple candidate recommendation information comes from multi-path recall, recall sources of the multi-path recall comprise at least one of selected from the group consisting of: user behavior of a user terminal matched with the current search information, social contact of the user terminal, and historical search mode, wherein data of the recall sources is stored in a storage device, and the multiple candidate recommendation information is obtained by accessing the storage device; determining, according to search volumes of the multiple candidate recommendation information within different search time limitations, sequence feature information of the candidate recommendation information set under each search time limitation, wherein the sequence feature information is used to reflect an overall search feature corresponding to the candidate recommendation information set; and determining target recommendation information based on a trained target ranking model and the sequence feature information of the candidate recommendation information set under each search time limitation; and displaying the target recommendation information on the GUI; wherein the determining target recommendation information based on the trained target ranking model and the sequence feature information of the candidate recommendation information set under each search time limitation, comprises: inputting the sequence feature information of the candidate recommendation information set under each search time limitation to the trained target ranking model that is run in the electronic device, to obtain an estimated click-through rate corresponding to each candidate recommendation information in the candidate recommendation information set; based on the estimated click-through rate corresponding to each candidate recommendation information, selecting a predetermined number of candidate recommendation information as the target recommendation information; wherein the trained target ranking model comprises a first ranking model and a second ranking model, and a model accuracy of the second ranking model is greater than a model accuracy of the first ranking model; inputting, by a terminal device, the sequence feature information of the candidate recommendation information set under each search time limitation to the trained target ranking model that is run in the terminal device, to obtain the estimated click-through rate corresponding to each candidate recommendation information in the candidate recommendation information set, comprises: selecting partial feature information in a feature information set comprising the sequence feature information of the candidate recommendation information set under each search time limitation, and inputting the selected partial feature information into the first ranking model running in the terminal device, and predicting an initial click-through rate of each candidate recommendation information; determining a first ranking result of the multiple candidate recommendation information according to the initial click-through rate corresponding to each candidate recommendation information; and determining, based on the first ranking result and the sequence feature information of the candidate recommendation information set under each search time limitation, the estimated click-through rate corresponding to each candidate recommendation information by using the second ranking model running in the terminal device.
  9. 9 . The electronic device according to claim 8 , wherein the sequence feature information comprises ranking feature information and variance feature information, the ranking feature information is used to reflect a ranking feature of the multiple candidate recommendation information in the candidate recommendation information set in terms of a size of search volume, and the variance feature information is used to reflect a deviation degree of the search volumes of the multiple candidate recommendation information with respect to an average search volume corresponding to the candidate recommendation information set.
  10. 10 . The electronic device according to claim 9 , wherein the variance feature information is determined according to steps comprising: for each search time limitation, determining an average search volume under the search time limitation according to the search volumes of the multiple candidate recommendation information in the candidate recommendation information set within the search time limitation; determining a search volume standard deviation under the search time limitation based on the average search volume and a search volume of each of the multiple candidate recommendation information within the search time limitation; and determining variance feature information of the candidate recommendation information set under the search time limitation based on the search volume standard deviation, the average search volume, and the search volumes of the multiple candidate recommendation information within the search time limitation.
  11. 11 . The electronic device according to claim 10 , wherein determining the variance feature information of the candidate recommendation information set under the search time limitation based on the search volume standard deviation, the average search volume, and the search volumes of the multiple candidate recommendation information within the search time limitation comprises: for each candidate recommendation information, determining a search volume difference value based on the average search volume and a search volume of the candidate recommendation information within the search time limitation, and dividing the search volume difference value by the search volume standard deviation to obtain a search volume ratio value of the candidate recommendation information; and determining the variance feature information of the candidate recommendation information set under the search time limitation based on search volume ratio values to which the multiple candidate recommendation information corresponds.
  12. 12 . The electronic device according to claim 8 , wherein determining the target recommendation information based on the trained target ranking model and the sequence feature information of the candidate recommendation information set under each search time limitation comprises: determining candidate feature information of each candidate recommendation information in the candidate recommendation information set, wherein the candidate feature information comprises text feature information and/or vertical category feature information; obtaining historical feature information authorized by a user, wherein the historical feature information comprises behavior feature information and/or background feature information; and determining, by using the trained target ranking model, the target recommendation information based on the historical feature information, the candidate feature information of each candidate recommendation information in the candidate recommendation information set, and the sequence feature information of the candidate recommendation information set under each search time limitation.
  13. 13 . The electronic device according to claim 12 , wherein the feature information set further comprises the historical feature information, and the candidate feature information of each candidate recommendation information in the candidate recommendation information set, and determining, by using the trained target ranking model, the target recommendation information based on the historical feature information, the candidate feature information of each candidate recommendation information in the candidate recommendation information set, and the sequence feature information of the candidate recommendation information set under each search time limitation comprises: selecting the partial feature information from the feature information set; determining, based on the partial feature information and the first ranking model, the first ranking result of the multiple candidate recommendation information in the candidate recommendation information set; determining, by using the second ranking model, a second ranking result of the multiple candidate recommendation information in the candidate recommendation information set based on the first ranking result, the historical feature information, the candidate feature information of each candidate recommendation information in the candidate recommendation information set, and the sequence feature information of the candidate recommendation information set under each search time limitation; and determining, according to the second ranking result, the target recommendation information from the candidate recommendation information set.
  14. 14 . The electronic device according to claim 13 , wherein determining, according to the second ranking result, the target recommendation information from the candidate recommendation information set comprises: selecting, according to ranked positions of the multiple candidate recommendation information as indicated by the second ranking result, a preset number of candidate recommendation information from the candidate recommendation information set, and determining the preset number of candidate recommendation information as the target recommendation information; and/or selecting, according to the estimated click-through rate of each candidate recommendation information as indicated by the second ranking result, candidate recommendation information with an estimated click-through rate greater than a threshold value from the candidate recommendation information set, and determining the candidate recommendation information with the estimated click-through rate greater than the threshold value as the target recommendation information.
  15. 15 . A non-transitory computer-readable storage medium, wherein the non-transitory computer-readable storage medium is configured to store computer-executable instructions, and the computer-executable instructions, when executed by a processor, cause the processor to: receive input information in a search box of an application program as current search information through a GUI (graphical user interface), and obtain a candidate recommendation information set which matches the current search information, wherein the candidate recommendation information set comprises multiple candidate recommendation information, the multiple candidate recommendation information comes from multi-path recall, recall sources of the multi-path recall comprise at least one of selected from the group consisting of: user behavior of a user terminal matched with the current search information, social contact of the user terminal, and historical search mode, wherein data of the recall sources is stored in a storage device, and the multiple candidate recommendation information is obtained by accessing the storage device; determine, according to search volumes of the multiple candidate recommendation information within different search time limitations, sequence feature information of the candidate recommendation information set under each search time limitation, wherein the sequence feature information is used to reflect an overall search feature corresponding to the candidate recommendation information set; and determine target recommendation information based on a trained target ranking model and the sequence feature information of the candidate recommendation information set under each search time limitation; and display the target recommendation information on the GUI; wherein the determine target recommendation information based on the trained target ranking model and the sequence feature information of the candidate recommendation information set under each search time limitation, comprises: inputting the sequence feature information of the candidate recommendation information set under each search time limitation to the trained target ranking model that is run in the processor, to obtain an estimated click-through rate corresponding to each candidate recommendation information in the candidate recommendation information set; based on the estimated click-through rate corresponding to each candidate recommendation information, selecting a predetermined number of candidate recommendation information as the target recommendation information; wherein the trained target ranking model comprises a first ranking model and a second ranking model, and a model accuracy of the second ranking model is greater than a model accuracy of the first ranking model; inputting, by a terminal device, the sequence feature information of the candidate recommendation information set under each search time limitation to the trained target ranking model that is run in the terminal device, to obtain the estimated click-through rate corresponding to each candidate recommendation information in the candidate recommendation information set, comprises: selecting partial feature information in a feature information set comprising the sequence feature information of the candidate recommendation information set under each search time limitation, and inputting the selected partial feature information into the first ranking model running in the terminal device, and predicting an initial click-through rate of each candidate recommendation information; determining a first ranking result of the multiple candidate recommendation information according to the initial click-through rate corresponding to each candidate recommendation information; determining, based on the first ranking result and the sequence feature information of the candidate recommendation information set under each search time limitation, the estimated click-through rate corresponding to each candidate recommendation information by using the second ranking model running in the terminal device.
  16. 16 . The medium according to claim 15 , wherein the sequence feature information comprises ranking feature information and variance feature information, the ranking feature information is used to reflect a ranking feature of the multiple candidate recommendation information in the candidate recommendation information set in terms of a size of search volume, and the variance feature information is used to reflect a deviation degree of the search volumes of the multiple candidate recommendation information with respect to an average search volume corresponding to the candidate recommendation information set.
  17. 17 . The medium according to claim 16 , wherein the variance feature information is determined according to steps comprising: for each search time limitation, determining an average search volume under the search time limitation according to search volumes of the multiple candidate recommendation information in the candidate recommendation information set within the search time limitation; determining a search volume standard deviation under the search time limitation based on the average search volume and a search volume of each of the multiple candidate recommendation information within the search time limitation; and determining variance feature information of the candidate recommendation information set under the search time limitation based on the search volume standard deviation, the average search volume, and the search volumes of the multiple candidate recommendation information within the search time limitation.
  18. 18 . The medium according to claim 17 , wherein determining the variance feature information of the candidate recommendation information set under the search time limitation based on the search volume standard deviation, the average search volume, and the search volumes of the multiple candidate recommendation information within the search time limitation comprises: for each candidate recommendation information, determining a search volume difference value based on the average search volume and a search volume of the candidate recommendation information within the search time limitation, and dividing the search volume difference value by the search volume standard deviation to obtain a search volume ratio value of the candidate recommendation information; and determining the variance feature information of the candidate recommendation information set under the search time limitation based on search volume ratio values to which the multiple candidate recommendation information corresponds.
  19. 19 . The medium according to claim 15 , wherein determining the target recommendation information based on the trained target ranking model and the sequence feature information of the candidate recommendation information set under each search time limitation comprises: determining candidate feature information of each candidate recommendation information in the candidate recommendation information set, wherein the candidate feature information comprises text feature information and/or vertical category feature information; obtaining historical feature information authorized by a user, wherein the historical feature information comprises behavior feature information and/or background feature information; and determining, by using the trained target ranking model, the target recommendation information based on the historical feature information, the candidate feature information of each candidate recommendation information in the candidate recommendation information set, and the sequence feature information of the candidate recommendation information set under each search time limitation.

Description

CROSS-REFERENCE TO RELATED APPLICATION The present application claims priority of the Chinese Patent Application No. 202310315196.2, filed on Mar. 28, 2023, the entire disclosure of which is incorporated herein by reference as part of the present application. TECHNICAL FIELD Embodiments of the present disclosure relate to the field of the Internet technology, in particular to an information recommendation method, an information recommendation apparatus, an electronic device, and a storage medium. BACKGROUND With the advancement of technology, more and more users acquire desired information through search means. For instance, a user may enter the search content within a search box to view search results related to the search content. For example, after the user enters the search content, search recommendation information related to the search content may be pushed for the user, so that the user may find desired information through the received search recommendation information. Thus, it can be seen that the precision of the search recommendation information may affect the search efficiency of the user. SUMMARY The embodiments of the present disclosure provide an information recommendation method, an information recommendation apparatus, an electronic device, and a storage medium. In a first aspect, at least one embodiment of the present disclosure provides an information recommendation method, and the information recommendation method includes: obtaining a candidate recommendation information set which matches current search information, the candidate recommendation information set including multiple candidate recommendation information; determining, according to search volumes of the multiple candidate recommendation information within different search time limitations, sequence feature information of the candidate recommendation information set under each search time limitation, the sequence feature information being used to reflect an overall search feature corresponding to the candidate recommendation information set; and determining target recommendation information based on a trained target ranking model and the sequence feature information of the candidate recommendation information set under each search time limitation. In a second aspect, at least one embodiment of the present disclosure further provides an information recommendation apparatus, and the information recommendation apparatus includes an obtaining module, a first determining module, and a second determining module. The obtaining module is configured to obtain a candidate recommendation information set which matches current search information, the candidate recommendation information set including multiple candidate recommendation information; the first determining module is configured to determine, according to search volumes of the multiple candidate recommendation information within different search time limitations, sequence feature information of the candidate recommendation information set under each search time limitation, the sequence feature information being used to reflect an overall search feature corresponding to the candidate recommendation information set; and a second determining module is configured to determine target recommendation information based on a trained target ranking model and the sequence feature information of the candidate recommendation information set under each search time limitation. In a third aspect, at least one embodiment of the present disclosure further provides an electronic device, and the electronic device includes a processor and a memory; the memory is in communication connection to the processor; one or more computer-executable instructions are stored on the memory; and the one or more computer-executable instructions, when executed by the processor, cause the processor to perform steps of the information recommendation method according to any one of the embodiments of the present disclosure. In a fourth aspect, at least one embodiment of the present disclosure further provides a computer-readable storage medium, the computer-readable storage medium is configured to store computer-executable instructions, and the computer-executable instructions, when executed by a processor, cause the processor to: obtain a candidate recommendation information set which matches current search information, the candidate recommendation information set including multiple candidate recommendation information; determine, according to search volumes of the multiple candidate recommendation information within different search time limitations, sequence feature information of the candidate recommendation information set under each search time limitation, the sequence feature information being used to reflect an overall search feature corresponding to the candidate recommendation information set; and determine target recommendation information based on a trained target ranking model and the sequence feature informa