Method, apparatus, computer device and storage medium for determining POI alias

Patent metadata

Field	Value
Publication number	US-11698261-B2
Application number	US-201916710566-A
Country	US
Kind code	B2
Filing date	Dec 11, 2019
Priority date	Apr 11, 2019
Publication date	Jul 11, 2023
Grant date	Jul 11, 2023

Abstract

Embodiments of the present disclosure provide a method, apparatus, computer device, and storage medium for determining a POI alias. The method may include: acquiring a to-be-processed target POI, and generating a candidate alias list corresponding to the target POI based on a query behavior-associated log matching the target POI; and screening out at least one target alias corresponding to the target POI in the candidate alias list, according to an association relationship between each candidate alias in the candidate alias list and the target POI.

First claim

What is claimed is:

1. A method for determining a point of interest (POI) alias, the method comprising:
acquiring a target POI that is to be processed, and generating a candidate alias list corresponding to the target POI based on a query behavior-associated log, the query behavior-associated log being used for recording data of query behavior directing at the target POI by a network user in an application providing location-based services (LBS), matching the target POI; and
screening out at least one target alias corresponding to the target POI in the candidate alias list, based on association relationships between candidate aliases in the candidate alias list and the target POI,
wherein before screening out at least one target alias corresponding to the target POI in the candidate alias list, the method further comprises:
extracting a sequence comprising at least one candidate alias and a standard name of the target POI from the query behavior-associated log, and building at least one alias-name pair from the sequence, the alias-name pair comprising a candidate alias and the standard name; and
eliminating a candidate alias consistent with the standard name of the target POI from the candidate alias list; or eliminating a candidate alias from the candidate alias list in response to a number of the alias-name pair of the candidate alias being less than a set number threshold, and
wherein the screening out at least one target alias corresponding to the target POI in the candidate alias list, according to association relationships between candidate aliases in the candidate alias list and the target POI, comprises:
acquiring at least one alias description feature corresponding to a candidate alias in the candidate alias list;
inputting the at least one alias description feature of the candidate alias into a pre-trained POI alias identification model, and acquiring an alias probability corresponding to the candidate alias output by the POI alias identification model; and
screening out the at least one target alias in the candidate alias list based on alias probabilities of the candidate aliases,
the pre-trained POI alias identification model being trained based on at least one piece of standard sample data obtained by:
based on query behavior-associated logs corresponding to a location service application, extracting a piece of sample group data comprising a user query formula, a user clicked POI, and a user actually visited POI for each query behavior-associated log, to obtain a plurality of pieces of sample group data corresponding to the query behavior-associated logs;
for each piece of sample group data, determining whether a user query formula thereof is different from a user clicked POI thereof and the user clicked POI thereof is identical with a user actually visited POI thereof, and determining the piece of piece target sample group to be a sample group in response to the user query formula thereof being different from a user clicked POI thereof and the user clicked POI thereof being identical with a user actually visited POI thereof;
constructing a piece of candidate sample data with the user query formula comprised in the target sample group as a candidate alias and a standard name of the user clicked POI as a standard name of a candidate sample POI;
acquiring word vectors of a candidate alias and the candidate sample POI of a piece of candidate sample data, and calculating a vector similarity between the candidate alias and the standard name of the corresponding candidate sample POI; and
screening out the standard sample data in the candidate sample data based on the vector similarities corresponding to respective pieces the candidate sample data.

2. The method according to claim 1 , wherein the generating a candidate alias list corresponding to the target POI based on a query behavior-associated log matching the target POI comprises at least one of:
in a map application, determining, from the query behavior-associated log, a first behavior log of navigating to the target POI selected by a user in a first query result page, and generating the candidate alias list based on a first query formula matching the first query result page in the first behavior log;
in a review application, determining, from the query behavior-associated log, a second behavior log of purchasing a service item corresponding to the target POI selected by a user in a second query result page, and generating the candidate alias list based on a second query formula matching the second query result page in the second behavior log; and
in a taxi application, determining, from the query behavior-associated log, a third behavior log of taking a taxi to the target POI selected by a user in a third query result page, and generating the candidate alias list based on a third query formula matching the third query result page in the third behavior log.

3. The method according to claim 2 , wherein the screening out at least one target alias corresponding to the target POI in the candidate alias list, according to association relationships between candidate aliases in the candidate alias list and the target POI, comprises:
acquiring, by a preset word vector model, word vectors of a standard name of the target POI and the candidate aliases in the candidate alias list;
calculating vector similarities between the standard name of the target POI and the respective candidate aliases; and
screening out the at least one target alias in the candidate alias list based on vector similarity calculation results.

4. The method according to claim 3 , wherein before acquiring, by a preset word vector model, word vectors of a standard name of the target POI and the candidate aliases in the candidate alias list, the method further comprises:
generating a text corpus based on at least one of review data or retrieval data matching the target POI;
performing word segmentation processing on a document comprised in the text corpus using a word segmentation dictionary, to obtain a plurality of segmented words, wherein the word segmentation dictionary comprises: the standard name of the target POI and the candidate aliases comprised in the candidate alias list; and
calculating, according to a preset word vector calculation method, word vectors of the segmented words as the word vector model.

5. The method according to claim 4 , wherein the generating a text corpus based on at least one of review data or retrieval data matching the target POI comprises at least one of:
in a review web site, acquiring a review page of the target POI, and composing comment data comprised in the review page into a document to generate the text corpus;
combining the first query formula in the first behavior log with the target POI into a document to generate the text corpus;
combining the second query formula in the second behavior log with the target POI into a document to generate the text corpus; and
combining the third query formula in the third behavior log with the target POI into a document to generate the text corpus.

6. The method according to claim 1 , wherein before acquiring at least one alias description feature corresponding to a candidate alias in the candidate alias list, the method further comprises:
acquiring at least one alias description feature respectively corresponding to the standard aliases in the respective pieces of standard sample data; and
inputting the at least one alias description feature respectively corresponding to the standard aliases in the respective pieces of sample data into a standard machine learning model, to train the standard machine learning model to obtain the POI alias identification model.

7. The method according to claim 6 , wherein the alias description feature comprises at le

Assignees

• Beijing Baidu Netcom Sci & Tech Co Ltd

Inventors

• Li Yanyan
• Duan Jianguo
• Xiong Hui

Classifications

• G06F40/242

  Dictionaries · CPC title

• G06N20/00

  Machine learning · CPC title

• G06F16/29

  Geographical information databases · CPC title

• G01C21/3476 Primary

  using point of interest [POI] information, e.g. a route passing visible POIs · CPC title

• G06N3/09

  Supervised learning · CPC title