Academic Journal of Science and Technology 
ISSN: 2771-3032 | Vol. 7, No. 1, 2023 

 

63 

Research	on	User	Profile	Combined	with	Collaborative	
Filtering	Recommendation	Algorithm	for	Intelligent	
Tourism	
Chun He, Caijian Hua 

School of Computer Science and Engineering, Sichuan University of Science & Engineering, Sichuan YiBin 643000, China 

 

Abstract: In recent years, the online travel sector in the tourism industry has experienced significant growth and popularity 
due to the development and widespread adoption of internet technology and smart devices. However, despite these advancements, 
scenic spots have struggled to provide precise services to tourists, as the online marketplace is flooded with numerous and 
disorganized commodity resources, lacking standardized construction and systematic management. As a result, travelers find it 
challenging to access specialty goods that cater to their personalized needs. To address this issue, this paper proposes the use of 
user profiling and collaborative filtering recommendation algorithms to achieve personalized recommendations for specialty 
products in scenic spots. The general process of constructing user profiles for scenic spots and combining them with collaborative 
filtering algorithms to create an intelligent tourism recommendation system is outlined. The paper also highlights the current 
challenges faced by this system in practical applications and provides future research prospects to promote accurate services in 
tourist attractions. 

Keywords: User profiling, Recommendation, Personalization, Tourism. 
 

1. Introduction 
In recent years, with the rapid development of the tourism 

industry, personalized travel service models and demands 
have been continuously strengthened, becoming a new focal 
point in the tourism market. Consumers are placing increasing 
importance on the quality of their travel experiences, seeking 
uniqueness and personalized adventures, and are willing to 
pay higher costs for them. Moreover, as the interest in travel 
grows among young people, they prefer independent and 
personalized travel styles, making personalized travel 
services and related products and offerings more popular. 
Personalized travel service is a consumer-oriented model, 
aiming to meet individualized needs by providing 
comprehensive, high-quality, and personalized tourism 
services through refined management and differentiated 
services. 

Consumers can choose travel routes, itineraries, and 
activities that suit their preferences while receiving 
professional advice and tailored services. This approach 
meets consumers' more personalized demands and offers a 
wider array of choices. However, the current scenic spots 
have not effectively addressed the personalized needs of 
travelers. Disorganized goods still flood major travel 
platforms, leaving tourists in a dilemma when selecting 
souvenirs and specialty products from scenic spots. To 
address this, in order to meet the increasingly urgent and 
efficient demand for intelligent travel information and 
effectively tackle the problem of information overload in the 
tourism industry, online travel service platforms urgently 
need to introduce intelligent and personalized travel 
recommendation services through continuous technological 
innovation. These services can provide users with more 
diverse and effective travel information and personalized 
recommendations for attractions, assisting users in decision-
making and itinerary planning. 

On April 6, 2023, the Ministry of Culture and Tourism of 

China issued the "Notice on Strengthening the Collaborative 
Innovation and Development of 5G+ Smart Tourism." This 
initiative was launched to deeply implement the "14th Five-
Year Plan for National Economic and Social Development 
and the Vision 2035," "14th Five-Year Plan for the 
Development of the Information and Communication 
Industry," "14th Five-Year Plan for the Development of the 
Tourism Industry," "5G Application 'Sailing' Action Plan 
(2021-2023)," and the "Opinions on Deepening 'Internet + 
Tourism' to Promote High-Quality Development of the 
Tourism Industry."The goal of this notice is to promote the 
innovative application of 5G technology in the tourism 
industry, encouraging market entities to explore 5G 
integration and innovation applications. The focus is on 
enhancing tourism services, improving the travel experience, 
innovating tourism management, optimizing tourism 
resources, and conducting business and model innovations.In 
line with this initiative, efforts will be made to nurture a group 
of 5G+ smart tourism solution providers and innovative 
enterprises. This will empower the scale development of 5G+ 
smart tourism innovative applications. Ultimately, this policy 
paves the way for personalized smart tourism, aiming to meet 
the individualized needs of travelers with advanced 5G-
powered solutions. 

User profiling refers to the process of collecting, 
organizing, and analyzing various data about users to 
construct comprehensive characteristics and descriptions, 
unveiling their interests, preferences, behavior patterns, and 
needs[1]. It serves as a fundamental cornerstone for 
personalized services and decision-making in 
recommendation systems, targeted advertising, and precision 
marketing. The primary objective is to assist businesses in 
gaining a better understanding of their users and providing 
more personalized and accurate products and services.On the 
other hand, collaborative filtering algorithms are a commonly 
used category of algorithms in recommendation systems[2]. 
They analyze user behavior and preferences to discover 



 

64 

similarities between users or items, thereby generating 
personalized recommendations for users. Built on the concept 
of "collective intelligence," collaborative filtering algorithms 
infer user behavior and preferences based on the shared 
behaviors of groups. Through the labels generated from user 
profiling, these algorithms can achieve personalized and 
precise product recommendations for tourist attractions[3]. 

In conclusion, this article aims to establish personalized 
tourism services with strong support from national policies. It 
lays a solid foundation for the personalized tourism 
ecosystem by employing user profiling techniques to 
construct user models and combining them with collaborative 
filtering algorithms to create a recommendation model for 
scenic spot specialties. 

2. Related Work 
In this chapter we discuss what is user portrait and the 

research status of smart tourism and collaborative filtering 
algorithm. 

2.1. User portrait overview 
Early user portraits were based on traditional market 

research and consumer behavior analysis. Marketers collect 
basic information and purchase behavior of users through 
telephone surveys, mailing questionnaires, etc. to understand 
user needs and preferences, and then use this information for 
advertising and product positioning[4]. 

With the development of the Internet and digital technology, 
user portraits have entered a new stage. The popularity of the 
Internet has made it possible to obtain a large amount of user 
behavior data, such as users' browsing records on websites, 
click behaviors, search habits, etc. These data provide more 
dimensional features for user portraits[5]. The rise of machine 
learning and big data technology has made the modeling and 
analysis of user portraits more efficient and accurate. By 
using various machine learning algorithms and models, users' 
interests and behavior patterns can be automatically learned 
from massive data, so as to construct user portraits more 
accurately. In recent years, the development of deep learning 
technology has further promoted the evolution of user 
portraits. Deep learning can process more complex user 
behavior data and text content, and extract deeper user 
characteristics, so as to achieve higher levels of personalized 
recommendation and precision marketing[6]. 

2.2. Research Status of Collaborative Filtering 
Algorithm 

In 1992, Paul Resnick and others proposed an early 
collaborative filtering algorithm and applied it to the 
recommendation of Usenet newsgroups. Based on the 
similarity between users, this method recommends items that 
are liked by similar users, thereby realizing personalized 
recommendation[7]. Subsequently, the collaborative filtering 
algorithm has received extensive attention and research, 
which has triggered an upsurge in the field of 
recommendation systems. Researchers have proposed various 
improvements and extensions, including item-based 
collaborative filtering, latent semantic models, matrix 
factorization, etc., to improve the accuracy and efficiency of 
recommendations. In the recommendation algorithm 
competition held by Netflix in 2006, the famous Matrix 
Factorization algorithm (Matrix Factorization) made a 

significant breakthrough in the collaborative filtering 
algorithm, making the collaborative filtering algorithm more 
mature and practical. 

In the application of collaborative filtering algorithm, Gao 
Chaomeng and Wang Yonggang[8] used collaborative 
filtering algorithm to analyze and study brand visual design 
in order to improve their own brand building level, and 
proposed a collaborative filtering algorithm to analyze the 
visual design of corporate brands due to the low accuracy of 
general recommendation. Communicate the design process. 
Quickly find the books you want, and propose a book 
recommendation algorithm based on collaborative filtering. 
Wang, Zhi Hui and Hou, De Zhi[9] take the interestingness of 
the book itself as an important measurement index, including 
search times, borrowing time, borrowing times, borrowing 
intervals, renewal times, etc. to improve the search rate. Wang, 
Tianyu and Ge, Dong[10] built a recommendation system 
based on a collaborative filtering multi-dimensional hybrid 
algorithm to find, analyze and recommend online Chinese 
learning resources that meet the needs of learners and teachers. 

2.3. Application of Collaborative Filtering 
Algorithm in Smart Tourism 

In terms of collaborative filtering applied to smart tourism, 
in order to find one’s favorite spots in a large number of 
tourist attractions, Niu, Tianjiao, Song, Mei, etc.[11] have 
improved the collaborative filtering algorithm based on the 
low accuracy of recommendation considering changes in user 
interests. The time function of Ebbinghaus' law of forgetting 
is used to calculate the user similarity to ensure the accuracy 
of the recommendation. Wang and Zhonghua[12] proposed 
that the intelligent recommendation model of tourist 
destinations requires an algorithm that can accurately 
recommend tourist attractions according to the interests of 
users. However, traditional collaborative filtering algorithms 
ignore the problem of user preferences, which may lead to a 
decrease in recommendation accuracy. To address these 
issues, he analyzed the factors that affect users' interest 
preferences based on their global and local rating information. 
By calculating the global probability distribution of user 
rating information, the Hamming approximation is used to 
calculate the user's interest preference. The similarity 
algorithm is effectively combined with the traditional 
similarity algorithm to propose a collaborative filtering 
recommendation algorithm model based on user preferences 
for tourist attractions under sparse data. 

3. Methodology 
In this chapter, we discuss the technical flow and design of 

user profile and collaborative filtering algorithm. 

3.1. The overall architecture of the model 
This paper proposes to build a model of user portraits to 

generate passenger tags, associate labels with items, extract 
corresponding tags or feature tags for the features of each 
scenic spot specialty, and calculate the similarity between 
user profile tags and scenic spot specialty labels to measure 
users' interest in goods, based on the similarity between user 
groups and scenic area specialty products. Select the group of 
users that are most similar to the target users to generate 
recommended columns for them. The conceptual model is 
shown in figure 1.： 

 



 

65 

 
Figure 1. Conceptual model diagram of user profile combined with collaborative filtering algorithm 

 

3.2. Construction of user portrait model 
The virtual user portrait is based on real data, and the result 

is a differentiated label description of different users after 
being segmented. User tags have semantic and short text 
features. The following mainly introduces the process of 
obtaining tags from passenger text data. Process the obtained 
scenic spot data through source data collection and data 
preprocessing, and complete the entire user portrait through 
the process of behavior modeling and user portrait creation. 

① Source data collection. Static data often comes from the 
information filled in by users on the website. When there is 
no data, it can be predicted by building a model. For example, 
when the user's gender information is unknown, a model can 
be built to judge based on the user's purchase behavior. 
Dynamic data refers to the data generated by user behavior, 
such as browsing scenic spots pages, purchasing scenic spot 
products, collections, evaluations, etc. These behavioral data 
can calculate the customer's brand preference, consumption 
ability, order quantity, category purchase ranking and other 

information. 
② data preprocessing. When the data is collected, it is 

generally necessary to do some preprocessing on the data, 
such as data cleaning and data structuring. While removing 
some dirty data, the structure of the data can also be 
standardized for better subsequent processing and analysis. 

③  Behavioral modeling. The user's static data and 
dynamic data are calculated through different dimensions of 
statistics and algorithm models, and the data is labeled to 
express the user's interests, needs, preferences, etc. 

④ User portrait creation. Deepen the behavior modeling 
process, and use prediction and Kmeans clustering algorithms 
to complete the mapping of different attributes of users to tags. 
The completion of user labeling means that the user's portrait 
is basically completed. 

Then build a user recommendation subsystem based on the 
data model of user portrait analysis to realize personalized 
recommendations for user products and scenic spots. 

The user portrait model architecture diagram of its smart 
tourism platform is shown in Figure 2: 

 

 
Figure 2. Architecture diagram of user portrait model 

 



 

66 

 

3.2.1. Chinese participle 
In the process of building user portraits, there are spaces 

between each English word as a very obvious separator[13], 
while Chinese words are continuous, which cannot be 
understood and processed by the computer. In order to make 
the Chinese text data generated by passengers into meaningful 
Words or phrases, so as to better understand and analyze user 
behavior and interests. This article will perform Chinese word 
segmentation processing on the obtained data, so that a piece 
of text can be divided into independent words. Commonly 
used Chinese word segmentation methods include dictionary-
based methods, statistics-based methods, and deep learning-
based methods. 

This article uses Jieba word segmentation as a tool for text 
processing. It combines word segmentation methods based on 
string matching and statistics. The word segmentation effect 
is better. The word segmentation process is as follows: 

1. Initialize the tokenizer. 
2. According to the characteristics of tourists, the scenic 

spot can customize the dictionary to contain specific words or 
terms. 

3. Use the forward maximum matching algorithm for word 
segmentation. The algorithm starts at the beginning of the text 
and matches the longest word. If a word is matched, the word 
will be used as a word segmentation result; if no word is 
matched, the current character will be output as a word. 

4. After the forward maximum matching word 
segmentation, process some rules and normalization, such as 
merging first and last names, removing single words, etc. 

5. The word segmentation is completed, and the stuttering 
word segmentation is used to output the results. 

3.2.2. Feature word extraction 
The main purpose of extracting feature words from user 

portraits is to convert passengers’ complex behaviors and 
interests into limited and meaningful features in order to 
better describe and understand users’ characteristics and 

behavior patterns[14]. Through feature word extraction, these 
behavioral data can be converted into a small number of 
meaningful feature words, thereby reducing the dimension 
and complexity of the data; it can accurately reflect the 
concerns of passengers, ignoring irrelevant or noise 
information; the characteristics of passengers and Behaviors 
are transformed into feature vectors, thereby realizing 
personalized recommendation of special products in scenic 
spots and calculation of passenger similarity. This paper uses 
TF-IDF as the main method of feature word extraction, counts 
the frequency of each word in the text, and multiplies the 
frequency of a word with the frequency of documents 
appearing in the text collection to get a more representative 
word The value of importance in the text. The calculation 
process of TF-IDF is as follows： 

(1). Calculate word frequency (TF) 
𝑻𝑭 𝑶𝒄𝒄𝒖𝒓𝒓𝒆𝒏𝒄𝒆𝒔 𝒐𝒇 𝒔𝒑𝒆𝒄𝒊𝒂𝒍 𝒘𝒐𝒓𝒅𝒔 

(2). Calculate inverse document frequency (IDF) 

𝑰𝑫𝑭 𝐥𝐨𝐠
𝑺𝑼𝑴

𝑵＋𝟏
 

Where SUM represents the total number of documents in 
the corpus, and N represents the number of documents 
containing the word 

(3). Calculate TF-IDF 
𝑇𝐹_𝐼𝐷𝐹 𝑇𝐹 𝐼𝐷𝐹 

3.2.3. label design 
The design of the user portrait label system can fully restore 

the user's characteristics, and improve the efficiency of 
computer analysis and processing of labels[15]. It is mainly 
divided into basic attribute tags and behavior feature tags. The 
basic attribute tags describe the basic information and 
characteristics of the user, and the behavior feature tags 
describe the user's behavior habits and behavior patterns. The 
user portrait label system of the smart tourism platform 
constructed in this paper is shown in Figure 3. 

 

 
Figure 3. Smart tourism platform user portrait label system 

 
Among them, the label represents the user's interest point, 

and the label weight is designed to represent the user's 
preference. According to TF-IDF, the user's weight value for 
the tag can be obtained, but the weight at this time only 
considers the relationship between the user and the tag, and 
the user's dynamic tag is constantly changing, so consider the 
business scene where the tag is located, the occurrence 
elements such as the time interval of the tag, the number of 
times the user generates the tag, etc. Therefore, this paper sets 
the user's dynamic label weight as the product of the behavior 
type weight (P), time decay (T), behavior times (N) and the 
weight (TF) obtained by TF-IDF. P focuses on the degree of 
collection of special products in scenic spots, T indicates the 
affected behavior, TF is the value obtained by Tf-IDF, and the 

greater the data of N, the greater the corresponding weight. At 
the end of time, according to the weight ranking of the 
obtained keywords, the nouns with the highest weight value 
are extracted as the user's feature labels, and the similarity 
between users can be obtained by calculating the label 
similarity between users (S). 

𝑆 𝑃 𝑇 𝑇𝐹 𝑁 

3.2.4. Cluster analysis 
The purpose of cluster analysis in the process of user 

portrait modeling is to group similar passengers into one 
category, thereby forming passenger groups or passenger 
clusters, and each group represents passengers with similar 
characteristics and behaviors. This facilitates the 
classification and segmentation of different travelers to better 

User Portrait tag system of 
Intelligent Tourism platform

Basic attribute tag
Behavior feature 

label

Age Gender Area
Educational 

level
...

Purchasing 
behavior

Search 
keywords

Browsing 
record

hobby ...



 

67 

understand the needs and behavior patterns of different 
groups of travelers. Common cluster analysis methods 
include K-Means Clustering, Density-Based Clustering, 
Spectral Clustering, etc. 

The text uses the K-Means Clustering method to cluster the 
passenger data, and iteratively finds K clusters to minimize 
the loss function corresponding to the clustering results. 
Among them, the loss function is defined as the sum of the 
squares of errors between each passenger sample and the 
center point of the cluster to which it belongs: 

𝐽 𝑐, 𝑢 ||𝑥 𝑢 ||  

Where 𝑥  represents the i-th passenger sample, 𝑐  is the 
cluster to which 𝑥  belongs, 𝑢  represents the center point 
corresponding to the cluster, and M is the total number of 
samples. The passenger clustering process is as follows: 

(1) Data preprocessing: standardization, outlier filtering. 
(2) Randomly select K centers, denoted as𝑢  , 𝑢  … 

𝑢 . 
(3) Define the loss function： 

𝐽 𝑐, 𝑢 ||𝑥 𝑢 ||  

(4) Let t=0,1,2,... be the number of iteration steps, repeat 
the following process until J converges: 

(4.1) For each sample 𝑥 , assign it to the nearest center 
𝑐 𝑎𝑟𝑔𝑚𝑖𝑛 ||𝑥 𝑢 ||  

(4.2) For each class center k, recalculate the center of the 
class 

𝑢 𝑎𝑟𝑔𝑚𝑖𝑛 ||𝑥 𝑢||
:

 

Finally, it is divided into K types of groups through cluster 
analysis. For each cluster, the commonality and 
characteristics of users can be analyzed, and one or more 
labels can be assigned to the cluster according to the designed 
label system. These tags can reflect the interests, behavior 
habits and other characteristics of the user group. Finally, the 
labels assigned to each cluster are associated with the original 
user data to construct a user group model. 

3.3. User portrait model combined with 
collaborative filtering algorithm 

The collaborative filtering algorithm can realize the 
personalization and accuracy of product recommendation 
through the tags generated by user portraits[16]. The tags in 
user portraits are descriptions of passenger interests, 
preferences, and behaviors, which can help collaborative 
filtering algorithms better understand users and make 
personalized recommendations. In this paper, each user group 
constructed by the user portrait is used as a data source, and a 
user-based collaborative filtering algorithm is used for 
recommendation. In the collaborative filtering algorithm, for 
the target passenger, find other users who are similar to them 
in the user portrait model, and according to the behavior and 
preferences of these similar users, recommend products that 
may be of interest to the target user and complete the 
personalized travel recommendation for passengers. The 
following is how the collaborative filtering algorithm uses the 
user group generated by the user portrait to realize the product 
recommendation process: 

(1). User group feature value extraction 
For each user group, group features are further extracted. 

In this paper, user age and passengers' different ratings on 
commodities are used as group features. These group features 

can help to better understand the commonality and needs of 
user groups. 

(2). Collaborative filtering algorithm 
When implementing personalized recommendation, user-

based collaborative filtering algorithm is used. For each user 
group, using the similarity calculation method, this paper uses 
the Pearson correlation coefficient to calculate the similarity, 
and find other user groups that are most similar to this group, 
that is, similar groups. The algorithm formula of the Pearson 
correlation coefficient is as follows: 

𝑟
∑ 𝑥 �̅� 𝑦 𝑦

∑ 𝑥 �̅� ∑ 𝑦 𝑦
 

Among them, 𝑥   takes the age of passengers as the 
behavior variable, �̅�  is the average age of passengers, 
𝑦 takes the ratings of passengers on different commodities as 
data variables, 𝑦 is the average rating of commodities, and n 
is the sample size of passenger group data. 

(3). Group recommendation 
Using the behavior and interests of similar groups as a 

reference, recommend items that may be of interest to the 
target user group. The recommendation results of similar 
groups can be weighted and averaged, or the recommendation 
results can be combined based on diversity. 

(4). Generate recommendation list 
According to the list of items recommended by the group, 

the final recommendation list is generated with the predicted 
score of the item as the weight, and presented to the target 
user group. 
The flow chart of its user portrait combined with 
collaborative filtering algorithm is shown in Figure 4： 

 

 
Figure 4. Flow chart of user portrait combined with 

collaborative filtering algorithm 
 

4. Result Analysis 
In this chapter, we compare the test standards and test 

results of the traditional collaborative filtering algorithm and 
the user portrait combined with the collaborative filtering 

Start

User profile modeling

Generate K user groups

Enter K1 (K1 nearest neighbor)

The similarity between user groups is calculated by Pearson, and the user 
group similarity matrix is obtained.

Ranking of user group 
matrix

Through the K1 nearest neighbor, calculate the matrix of the interest 
degree of the user group to the item.

The top N are recommended by the degree of interest in 
the items.

Calculate recall, accuracy and 
coverage

End



 

68 

algorithm. 
The collaborative filtering recommendation algorithm 

combined with user portraits is aimed at a large number of 
mobile users in scenic spots, and solves the problem that it is 
difficult to make accurate predictions and recommendations 
due to the lack of behavior data of new tourists. In this paper, 
new users are divided into user groups, and the 
recommendation list constructed based on the behavior 
characteristics and attribute characteristics of old users is used 
as the basis to achieve the purpose of accurate and 
personalized recommendation of special products in scenic 
spots for new and old users of scenic spots. In order to ensure 
the authenticity and scientificity of the data, this paper 
collects real data based on tourism websites as the old user 
data set, and manually adds new user passenger information 
as the new user data set. 

4.1. Measured standard 
There are three evaluation indicators in the smart travel 

recommendation system, which are accuracy, novelty and 
coverage[17]. Accuracy is a test of the results of an algorithm. 
Accuracy is used as the index to evaluate the experiment. The 
experiment uses MAE as the evaluation standard for test 
verification. MAE reflects the gap between the actual rating 
value and the predicted rating value[18]. The smaller the gap 
between the two, the better the recommendation effect. It is 
assumed that the predicted rating set of the user portrait 
combined with the collaborative filtering model is 
{p1,p2,...,pn}, the user group's rating set for the product is 
{q1,q2,...,qn}, the calculation formula of MAE can be 
expressed as： 

𝑀𝐴𝐸
∑ |𝑝 𝑞 |

𝑁
 

Among them, N is the size of the test set, 𝑝   is the 
predicted rating value given by the recommendation 
algorithm, and 𝑞   is the actual average rating of the i 
commodity by the actual passenger group. The larger the error 
of MAE, the larger the value. The smaller the value of MAE, 
the better the accuracy of the prediction model. 

4.2. Result analysis 
The results of the test set are shown in Figure 5. Under the 

condition of dividing the scale of passenger set the number of 
groups=10, 20,...,60, it compares the accuracy of the 
traditional collaborative filtering algorithm when testing the 
same number of new passengers. 
 

 
Figure 5. Comparison of algorithm test set results 

 
It can be seen from the change trend of MAE in the figure 

that the MAE of the two algorithms will change with the 
number of neighbors of the passenger group. When the 
number of neighbors is small, because only a few people’s 
opinions are referred to, the prediction score is greatly 
affected by personal factors. Compared with passengers who 
only have attribute characteristics, the reference value of 
neighbors is small, resulting in their MAE Larger, the 
prediction accuracy is lacking. The collaborative filtering 
algorithm combined with user portraits is grouped under 
similar attribute features, based on the behavior 
characteristics of old users, and by referring to the data of the 
neighbors of the user group, the more the number of 
neighbors of the passenger group within a certain range, the 
smaller the MAE, and the better the prediction results. 
accurate. From the above analysis, the collaborative filtering 
algorithm combined with user portraits proposed in this paper 
is superior to traditional collaborative filtering algorithms in 
terms of recommendation performance for new passengers 
and users without behavioral characteristics. 

5. Conclusion and Discussion 
Travel recommendation is a specific application field of 

intelligent recommendation system, and collaborative 
filtering recommendation technology is the most commonly 
used recommendation algorithm. This paper proposes to 
generate user tags by building a model of user portraits, and 
associate tags with items. For the feature extraction of each 
scenic spot specialty, I want to extract the corresponding tags 
or feature tags, and calculate the similarity between the tags 
in the user portrait and the scenic spot specialty tags. , to 
measure the user's interest in the item, based on the similarity 
between the user and the special products of the scenic spot, 
select a group of users who are most similar to the target user 
to generate a recommendation list for it, so as to solve the 
problem of excessive user base in collaborative filtering 
recommendation sparsity problem. 

In this way, the collaborative filtering algorithm uses the 
tags generated by user portraits to achieve personalized and 
accurate product recommendations. It can better understand 
the interests and needs of passengers, so as to provide 
passengers with product recommendations that meet their 
personalized preferences, and improve passenger satisfaction 
and purchase experience. At the same time, based on the 
findings of this paper, there are still some directions for 
improvement in the future: 

1. Solve the cold start problem: 
When a new user joins the system or a new item is launched, 

due to the lack of user behavior data or item rating 
information, traditional collaborative filtering algorithms and 
user portraits may not be able to make accurate 
recommendations. Solving the cold start problem requires 
introducing other information sources, such as content 
information, social network data, etc., to help establish 
portraits and recommendation models for new users or new 
items. 

2. You can add context information: 
The current user portrait combined with collaborative 

filtering algorithm is mainly based on the user's historical 
behavior and characteristics, but ignores the contextual 
information when recommending, such as time, geographical 
location, device, etc. Integrating contextual information into 
the recommendation model can better adapt to the current 
needs and situations of users and provide more accurate 
recommendations. 



 

69 

3. By integrating multiple recommendation algorithms: 
Although the collaborative filtering algorithm performs 

well in personalized recommendation, it is not suitable for all 
scenarios. Collaborative filtering algorithm can be integrated 
with other recommendation algorithms (such as content 
recommendation, deep learning model, etc.) to form a hybrid 
recommendation system to make full use of the advantages of 
different algorithms and improve the diversity and accuracy 
of recommendations 

4. User privacy protection: 
User portraits involve the user's personal information and 

behavioral data, and it is necessary to protect the user's 
privacy and data security. In the recommendation model, 
privacy protection measures should be strengthened, such as 
adding a blockchain network architecture to ensure data 
security and prevent user portraits from being abused or 
leaked. 

References 
[1] Eke C I, Norman A A, Shuib L, et al. A survey of user profiling: 

State-of-the-art, challenges, and solutions[J]. IEEE Access, 
2019, 7: 144907-144924. 

[2] Zhu L, Li H, Feng Y. Research on big data mining based on 
improved parallel collaborative filtering algorithm[J]. Cluster 
Computing, 2019, 22: 3595-3604. 

[3] Hong B, Yu M. A collaborative filtering algorithm based on 
correlation coefficient[J]. Neural Computing and Applications, 
2019, 31: 8317-8326. 

[4] Yi X, Bertino E, Rao F Y, et al. Privacy-preserving user profile 
matching in social networks[J]. IEEE Transactions on 
Knowledge and Data Engineering, 2019, 32(8): 1572-1585. 

[5] Zhao S, Li S, Ramos J, et al. User profiling from their use of 
smartphone applications: A survey[J]. Pervasive and Mobile 
Computing, 2019, 59: 101052. 

[6] Ouaftouh S, Zellou A, Idri A. Social recommendation: A user 
profile clustering‐based approach[J]. Concurrency and 
Computation: Practice and Experience, 2019, 31(20): e5330. 

[7] Zhu L, Li H, Feng Y. Research on big data mining based on 
improved parallel collaborative filtering algorithm[J]. Cluster 
Computing, 2019, 22: 3595-3604. 

[8] Chaomeng G, Yonggang W. Analysis of Brand Visual Design 
Based on Collaborative Filtering Algorithm[J]. Discrete 
Dynamics in Nature and Society, 2022, 2022: 1-8. 

[9] Wang Z H, Hou D Z. Research on book recommendation 
algorithm based on collaborative filtering and interest degree[J]. 
Wireless Communications and Mobile Computing, 2021, 2021: 
1-7. 

[10] Wang T, Ge D. Research on Recommendation System of 
Online Chinese Learning Resources Based on Multiple 
Collaborative Filtering Algorithms (RSOCLR)[J]. 
International Journal of Human–Computer Interaction, 2023: 
1-11. 

[11] Niu T, Song M, Wang X, et al. Tourism Destination 
Recommendation and Marketing Model Analysis Based on 
Collaborative Filtering Algorithm[J]. Mobile Information 
Systems, 2022, 2022. 

[12] Wang Z. Intelligent recommendation model of tourist places 
based on collaborative filtering and user preferences[J]. 
Applied Artificial Intelligence, 2023, 37(1): 2203574. 

[13] Wang H, Tu Z, Fu Y, et al. Time-aware user profiling from 
personal service ecosystem[J]. Neural Computing and 
Applications, 2021, 33: 3597-3619. 

[14] Sharma S, Gupta V. Role of twitter user profile features in 
retweet prediction for big data streams[J]. Multimedia Tools 
and Applications, 2022, 81(19): 27309-27338. 

[15] Singh P K, Othman E, Ahmed R, et al. Optimized 
recommendations by user profiling using apriori algorithm[J]. 
Applied Soft Computing, 2021, 106: 107272. 

[16] Jin Q, Zhang Y, Cai W, et al. A new similarity computing 
model of collaborative filtering[J]. IEEE Access, 2020, 8: 
17594-17604. 

[17] Li K, Zhou X, Lin F, et al. Deep probabilistic matrix 
factorization framework for online collaborative filtering[J]. 
IEEE Access, 2019, 7: 56117-56128. 

[18] Zhang B, Zhang H, Sun X, et al. Integrating an attention 
mechanism and convolution collaborative filtering for 
document context-aware rating prediction[J]. IEEE Access, 
2018, 7: 3826-3835