A Point of Interest Intelligent Search Method based on Browsing
History
Wei Sun
1
, Yang Cong
1
, Xiaoli Liu
2
and Chengming Li
2
1
College of Geomatics, Shandong University of Science and Technology, Qingdao, China
2
Chinese Academy of Surveying and Mapping, Beijing, China
Keywords: Point of Interest Search Method, Spatial Heat and Scoring Algorithm.
Abstract: Point of interest search is one of the core search functions in GIS. The conventional methods may not be able
to personalize search results because the user’s personal interests are not taken into account. In order to
address this problem, this article proposes a point of interest intelligent search method based on the user’s
browsing history of map tiles. First, analyze the user’s map tile browsing history data on a smart city platform
and visualize the user’s focal hotspots through a heat map to derive the spatial heat. Second, add the spatial
heat influence factor to the attribute query to influence the search results of point of interest, so that the results
are more consistent with the user's search intention and corresponding services are offered to different users
based on their hotspots. Finally, an experiment with point of interest data from the City of Tengzhou verifies
the effectiveness and advantage of the method proposed by this article.
1 INTRODUCTION
Point of interest search is one of the core search
functions in Geographic Information System (GIS),
and the accuracy of the search results is directly
related to the practicability of the system. Many
scholars have done in-depth research on point of
interest search. Zhong (2007) proposed a rule-based
point of interest search algorithm, but it can lead to
errors if there is an irregular full name in the database.
Mao (2008) proposed a point of interest Chinese
abbreviation retrieval algorithm based on statistical
model. This algorithm does not have the full names
based on segmentation information and does not
discern the user’s characteristics either, so the
accuracy of matching is low. Cen (2010) proposed a
method of reordering the search results, which uses
log information to construct a collection of the user’s
high frequency words. When the user submits a query
request, it combines the content of the collection and
search results, and presents reordered result web
pages to the user. Miao (2012) proposed a text
filtering system based on vector space model (VSM).
In the field of GIS, point of interest data accumulate
over time, and a large amount of data will lead to
increasingly reduced search efficiency. Therefore, a
search method is needed to extract the best matching
results from the massive amount of data sources and
push them to the users based on their interests to
improve the search efficiency and make it more
intelligent. However, these algorithms do not achieve
the effect of intelligent search of point of interest.
Therefore, this paper proposes an intelligent search
method of point of interest applicable for GIS based
on the user’s map tile browsing history.
2 GENERAL SEARCH METHODS
In the existing research, the typical point of interest
search method is a string-based word segmentation
matching method. This process is shown in Figure 1.
First, the search keyword for the address is
segmented, as the acquisition of point of interest
requires search keyword segmentation and semantic
normalization. Second, based on a specific strategy,
the segmented string and the entries in the machine
dictionary are matched, and the relevance score of the
search results are calculated according to the degree
of matching. Finally, the search results are obtained
based on the order of the relevance score, and the
position point coordinates are matched based on the
results.
Sun, W., Cong, Y., Liu, X. and Li, C.
A Point of Interest Intelligent Search Method based on Browsing History.
DOI: 10.5220/0007712301890192
In Proceedings of the 5th International Conference on Geographical Information Systems Theory, Applications and Management (GISTAM 2019), pages 189-192
ISBN: 978-989-758-371-1
Copyright
c
2019 by SCITEPRESS – Science and Technology Publications, Lda. All rights reserved
189
Figure 1: String-based word segmentation matching method.
String-based word segmentation matching
method does not take user's interests into account. It
only matches the word segmentation results to the
database and simply returns the search results. This
process leads to the same search results for different
user groups if they are using the same keywords.
Therefore, understanding and managing user ’ s
needs and an active push require a higher standard for
the search method.
3 POINT OF INTEREST
INTELLIGENT SEARCH
METHOD BASED ON USER'S
MAP TILE BROWSING
HISTORY
3.1 Spatial Heat Calculation
Spatial heat refers to the degree of user’s attention
to a region. A high spatial heat in an area indicates the
area is likely a focal area for the user. In this article,
spatial heat is calculated using the user's map tile
browsing history.
1) The first step is to obtain and filter the log data in
order to get the Web Map Tile Service (WMTS) path
and number of calls invoked by the user. Log
collection is performed using Filebeat. Elastic.co
provides the most comprehensive support for Filebeat
and easy access to Elasticsearch, the widely used
analysis and retrieval system. Filebeat has a lower
resource overhead compared to Logstash. It is
characterized by minimal amount of coding and easy
optimization. Therefore, Filebeat has better
performance compared to the traditional log
acquisition methods.
2) There is information redundancy in the data
collected by Filebeat, and the search results need to
be filtered again in Elasticsearch to eliminate invalid
data. Elasticsearch is a distributed storage system
based on full-text indexing. It expands and
encapsulates the Lucene service engine with a user
interface, so the operation commands can be called
directly by the user to complete a specified task. In
addition, the full-text indexing in Elasticsearch can
create segments and shards of multiple sizes. It can
also create several shard replicas stored on each data
node of the Elasticsearch cluster. Meanwhile, it also
provides the ability to process, distribute, and assist
these shards and replicas, ensuring organized,
balanced, and automatically routed communication
among the nodes.
3) Using Elasticsearch to parse the service path can
obtain the tile specific location, service resolution,
range, and coordinates of starting point. Analyzing
the map tile browsing history can yield tile names,
paths, zoom levels, row and column numbers, and
number of calls. Based on this information, the
coordinates of the corresponding tiles are calculated.
The region information is obtained by matching the
coordinates to the database. User’s spatial heat is
generated based on the number of calls and the name
of the location of the map tiles. The formula for
calculating geographical coordinates is as follows:
（
(1)
（
）
(2)
In Formula (1) and (2), lon is the center point
longitude, lat is the center point latitude, x is the tile
row number, y is the tile column number, res is the
resolution level, theight is the tile length, twidth is the
tile width, XOrigin is the x axis starting point, and
YOrigin is the y axis starting point.
3.2 Scoring Algorithm and
Optimization
After matching to a set of documents, they need to be
sorted by the level of relevance. Not all documents
contain all the query words, and some words are more
important than others. The relevance score of a
document partially depends on the weight of each
query word in the document. The commonly used
scoring algorithm employs the Boolean model to find
matching documents and calculates the relevance
using the practical scoring function. In this article, the
GISTAM 2019 - 5th International Conference on Geographical Information Systems Theory, Applications and Management
190
spatial heat influencing factor is added to optimize the
common algorithm. The formula is as follows:
(3)
(4)
(5)
(6)
(7)
(8)
Formula (3) is the optimized scoring algorithm,
and Formula (4) is the common scoring algorithm.
In Formula(4), coor is the coordination
factor, which is based on the number of query items
contained in the document. This factor weights
documents that contain more query items using a
method similar to AND.
is the normalized value for each
search, which is the square sum of the weights of each
query item. The shorter the field, the higher the
weight of the field. If a word appears in a field such
as the title, it is more relevant than when it appears in
a field such as the body of content.
is the inverse document frequency (idf),
which is the frequency of a term appearing in all
documents in the collection. The higher the
frequency, the lower the weight. It is used to measure
the "uniqueness" of an item. Terms with a higher
frequency has a lower idf, and uncommon terms have
a higher idf. The inverse document frequency is the
logarithm of the number of documents in the index,
divided by the number of documents that contain the
term, the calculation is as Formula(5).
is the frequency of a term
appearing in the document. The higher the frequency,
the higher the weight. A field that mentions the same
term five times is more relevant than a field that
mentions it only once. Term frequency is the square
root of the number of times the term appears in the
document, the calculation is as Formula(6).
is the weight of query item.
is a weighted factor related to length.
In Formula(3), num
is the number of calls.
is the level of matching between the keywords
and the spatial heat, the calculation is as Formula(7).
In Formula(7), overlap is the number of matched
terms in the query, maxoverlap is the total number of
terms in the query.
is the weight of the spatial heat, the calculation
is as Formula(8). In Formula(8),
is the number of
calls of the map tile k,
is the total number of calls
for all tiles.
4 EXPERIMENT AND ANALYSIS
4.1 Environment and Data
In this article, the experimental data was the point of
interest data from Tengzhou, Zaozhuang, Shandong
Province. The index database was constructed using
Elasticsearch. The search framework was constructed
using front-end web technology. The experimental
data was analyzed to empirically verify the method
proposed by this article. The steps are as follows:
Step 1: Use Elasticsearch to retrieve the users’
map tile call log data.
Step 2: Filter the search results: Retain user
records with a Tilematrix value less than 13.
Step 3: Obtain the Tilematrix, Tilecol, TileRow of
the called map tile and calculate the center point
coordinates and range of the tile.
Step 4: Visualize data to form images, calculate
and , and add them to the score sorting of point of
interest search results.
Step 5: Compare the optimized point of interest
search method to the original method. Carry out
experiments with 6 different search terms, and each
term is searched 50 times. The returned results are
compared, and the corresponding precision ratios are
calculated.
4.2 Results and Analysis
Table 1 shows the comparison of experimental data
between the improved and the original
algorithms.After comparing the results between the
two methods, the improved method had a similar
precision ratio with the original algorithm when the
log data was empty. However, along with the
accumulation of the user's map tile browsing history
data, the outcome of the optimized search method was
significantly improved, whereas the accuracy of the
original method was basically unchanged. It
demonstrated that the method proposed in this article
can return the maximum relevant results based on the
user’s focal hotspot areas, which makes the search
more personalized and intelligent.
A Point of Interest Intelligent Search Method based on Browsing History
191
Table 1: Comparison of experimental data between the improved and the original algorithms.
Number
Search Term
The 1st Search
(Precision of
the Improved
Algorithm)
The 50th Search
(Precision of the
Improved
Algorithm)
The 1st Search
(Precision of the
Original
Algorithm)
The 50th Search
(Precision of the
Original Algorithm)
1
Gas Station
0.782
0.832
0.782
0.782
2
Sinopec Gas
Station
0.860
0.942
0.857
0.857
3
Police Station
0.754
0.864
0.755
0.755
4
Kindergarten
0.673
0.713
0.672
0.672
5
First Hospital
0.812
0.957
0.812
0.812
6
Dongshahe
Hospital
0.946
0.987
0.946
0.946
5 CONCLUSIONS
This article proposes a point of interest intelligent
search method based on the user's map tile browsing
history, which calculates the user’s focal spatial heat
through obtaining and analyzing the user's map tile
browsing history and uses the spatial heat as an
influence factor to optimize the scoring algorithm.
Through the comparison with the original algorithms,
the following conclusions can be drawn:
(1) The user’s log data contains information about
the interests of the web user. The research based on
log data can help us understand and master the real
needs of web users. The map tile search records can
profile the characteristics of the user's interests.
(2) In the field of GIS, using the map tile search
records can achieve intelligent search of point of
interest. This article studies the user’s map tile
browsing history and proposes an optimized
algorithm to add spatial heat to the search method.
Through the comparison and analysis of simulation
experiments, this improved method can effectively
improve the precision ratio of search, and become
more intelligent with the increase of data volume.
This article modifies the search results based on
the user's spatial heat. However, the in-depth research
on the search mechanism and user’s interest model
is still lacking. The next steps will involve studying
the mechanism and structure of search algorithms and
formation and application of user models, in order to
improve the quality of point of interest search and
efficiency of personalized information services..
ACKNOWLEDGEMENTS
This research was funded by a project supported by
the National Key Research and Development
Program of China (Grant No. 2016YFF0201305) and
National Natural Science Foundation of China(Grant
No. 41871375).
REFERENCES
Ceng R, Liu Y, Zhang M. 2010. Search Engine User
Behavior Analysis Based on Log Mining. Journal of
Chinese Information Processing. 24(3): 49-54.
Youdong M, Xipeng Qiu, Xuanjing H. 2012. A FAST
ALGORITHM FOR LARGE SCALE WEB PAGE
CLASSIFICATION. Computer Applications &
Software, 29(7):260-263.
Zhujie M, Xuehu Z. 2008. Study of Chinese Points of
Interest Search Based on Name Abbreviations. Journal
of Taiyuan University of Technology, 39:52-55.
Zhong L, Zheng F. 2007. Study on Approaches to Retrieval
of Chinese Organization Name Based on its
Abbreviated Name. Journal of Chinese Information
Processing. 21(1): 38-42.
GISTAM 2019 - 5th International Conference on Geographical Information Systems Theory, Applications and Management
192
