=Paper=
{{Paper
|id=Vol-3143/paper5
|storemode=property
|title=Information Overload and User Satisfaction: Balance Between Reliance on Recommendations and Deliberate News Selection
|pdfUrl=https://ceur-ws.org/Vol-3143/paper5.pdf
|volume=Vol-3143
|authors=Zhixin Pu,Michael A. Beam
|dblpUrl=https://dblp.org/rec/conf/recsys/PuB21
}}
==Information Overload and User Satisfaction: Balance Between Reliance on Recommendations and Deliberate News Selection==
https://ceur-ws.org/Vol-3143/paper5.pdf
Information Overload and User Satisfaction: Balance Between
Reliance on Recommendations and Deliberate News Selection
Zhixin Pu 1, Michael A. Beam 2
1
The Pennsylvania State University, 8 Carnegie Building, State College, United States
2
Kent State University, 334 Franklin Hall, Kent, United States
Abstract
High content relevance of news recommendations is a key factor for personalized user
experiences. Personal factors, such as seeking/scanning and goal commitment, also impact
information overload and user satisfaction in news recommender systems. This experimental
study using data from 669 Amazon MTurk workers tests a theoretical process of news selection
in news recommender systems. We manipulated two key elements of the news
recommendation system: relevance of news article recommendations and the presence of a
search bar. Results indicated that recommendation systems providing more irrelevant news
recommendations results in users selecting more irrelevant articles and reporting higher
information overload and lower satisfaction. Though we did not find evidence that seeking
news with a search bar would positively influence goal commitment compared with those
scanning a news recommendation portal, goal commitment significantly impacts news
selection and perceived information overload and user satisfaction.
Keywords 1
Information overload, news selection, personalization, news recommender systems, user
satisfaction, goal commitment
1. Introduction
The internet opens access to a flood of news information to users, leading to difficulties of news
information processing and decision making. Recommendation systems and personal news portals are
frequently used sources of health-related news, such as Google searching. These systems reorder and
prioritize news information to maximize relevance to the user [1, 2]. The pre-selection of information
in recommendation systems alters traditional information processing process via algorithm-selected
news information, which may reduce information overload. Research on recommendation systems in
different contexts has resulted in contradictory findings relative to reduced information overload [3, 4].
For example, recommendation systems can offer users a way to seek specific information using a search
bar or scan for specific information after presenting several linked items. The potentially differential
impact of information seeking and scanning has not been investigated relative to information overload.
This study aims to compare seeking and scanning to clarify the effectiveness and elucidate the
theoretical process of recommendation systems on information processing.
Algorithmically personalized systems prioritize the most relevant information [1,2] and have been
related to information overload reduction [5,6]. However, other work on personalized recommendation
systems has found overall relevance on SNSs [7] or personalized news portals [8] as not significantly
predictive of information overload. Although past studies have indicated the benefits of relevant
recommendations on combating information overload, the relationship between recommendations and
INRA'21: 9th International Workshop on News Recommendation and Analytics, September 25, 2021, Amsterdam, Netherlands
EMAIL: zmp5110@psu.edu (A. 1); mbeam6@kent.edu (A. 2).
ORCID: 0000-0002-2293-0845 (A. 1); 0000-0003-2425-2010 (A. 2);
©️ 2021 Copyright for this paper by its authors.
Use permitted under Creative Commons License Attribution 4.0 International (CC BY 4.0).
CEUR Workshop Proceedings (CEUR-WS.org)
the information selection process remain understudied. This paper also explores how the relevance of
recommendations impacts information overload and user satisfaction through users’ news selection
choices in the context of two common information behaviors: information scanning and seeking.
Although high relevance of recommendations decreases information overload, users still experience
difficulties selecting certain articles to read when facing several related-article recommendations [9].
Relying on the ranking of recommender systems may mitigate choice overload. However, the efficacy
of recommender systems and users’ reliance on the recommender systems to help select relevant articles
for decision making is under studied. Our study offers a 2x2 experiment where we manipulated the
relevance of recommendations and whether users are scanning or seeking in the recommendation
system. We rely on two metrics of selection behaviors. First, we measured the number of articles
selected by users at the top of the recommended list. Second, we measured the number of irrelevant
articles selected by users. Taken together, we use these selection metrics to explore to what extent users
rely on the recommendation system for filtering. In addition, we argue that goal commitment is a key
individual difference that should impact information overload.
In sum, this study adds to the literature by illustrating the process of how the efficacy of news
recommendation systems impact news selection, which in turn affects information overload and user
satisfaction. We explore how this process differs in the context of news seeking and news scanning.
The study enhances our theoretical understanding of how people select and consume information with
the aid of filtering recommendation systems with quantitative evidence about the impact of relevance
in recommender systems on users’ news consumption. The study also provides practical suggestions
regarding leveraging recommender systems efficiently. The study also indicates the importance of
relevance and goal commitment in the design of news recommender systems.
2. Literature Review
2.1 Information Overload
Scholars have not reached a consensus on a universal agreed definition of information overload [10,
11,12]. Some scholars refer information overload as exposure to diverse issues [13, 14, 15]. For example,
Klapp [16] described the concept as the rate of information that is too high for individuals to digest,
which triggers distraction, stress, errors in information processing stage. Hiltz and Turoff [17] claimed
people are overloaded without the capability of filtering useful information among a plethora of
information. One consistent explanation of information overload is that people are overwhelmed when
they can’t process excess information. Thus, we define information overload as the sensation
experienced when people encounter excess information that they cannot process.
Prior researches have shown that several variables trigger information overload including
demographics [18], levels of internet use [19], frequency of news access, level of attention to news, and
interest in news [18].
2.2 Relevance, news selection, and information overload
Relevance refers to any pattern that attracts our attention and requires interpretation by
individuals .We define relevance in recommender systems similar to the definition proposed by Pazzani
et al. [20], which refers to the information vector between individuals’ topic interests and suggested
news items. Relevance of an item is widely discussed and applied in the design of news recommender
systems to predict and offer useful and accurate content to users. Collaborative filtering [21] and content
filtering were approaches to recommend relevant news articles to users.
Another common technique for filtering is “top N” recommendations, which aims to provide a small
set of relevant articles to each user [22]. Content listed in the “top N” position was selected based on a
highest predicted ratings approach [23], and was predicted to be the ones that are be of interest to users
and are more likely to be clicked by users. Relevant recommendations were believed to assist with news
selection process. Top articles and relevant articles are the ones that are predicted to receive high click
rates. Relevant recommendations were also believed to reduce information overload and increase user
satisfaction. That is to say, relevance impacts news selection. Highly relevant recommendations predict
more top articles and less irrelevant articles selected by users. News selection further impacts
information overload and user satisfaction. More top articles and less irrelevant articles selected by
users predict lower information overload and lower user satisfaction in high relevance condition, while
the situation might not hold true in low relevance condition. Thus, we propose that:
H1: Highly relevant recommendations are a) negatively correlated with the number of
irrelevant articles read, and b) positively correlated with the number of top articles read.
H2: More irrelevant articles read by users are correlated with a) higher information overload
and b) lower user satisfaction.
H3: More top articles read by users are correlated with a) lower information overload and b)
higher satisfaction for those who receive more relevant recommendations.
H4: High relevance of recommendations is directly correlated with a) lower information
overload and b) higher user satisfaction.
H5: High relevance of recommendations is indirectly correlated with a) lower information
overload and b) higher user satisfaction.
2.3 Scanning/Seeking, news selection, and information overload
Seeking information and scanning information are two actions that are related to one basic need [24].
Information seeking, defined as goal-oriented information searching [24], differs from information
scanning when individuals may read information without clear information seeking purposes in
opportunistic browsing [25]. Existing research about news recommender systems rarely explores the
difference between scanning and seeking on news selection. Though browsing articles on a certain topic
and searching for information might not determine levels of information overload, the goal commitment
of finding answers for questions might influence news selection, further impacting information overload
and user satisfaction. Uses and gratifications theory [26] illustrated that users choose specific media to
satisfy their needs. Users are satisfied when their information need is met by finding information on the
recommender systems. High goal commitment is more likely to increase user satisfaction and decrease
information overload when the information needs are met. However, low goal commitment indicated
low information needs and thus may not lead to higher user satisfaction or impact information overload.
In addition, news selection mediates users’ satisfaction and information overload. People with high
goal commitment are more determined to find certain information and thus may read fewer irrelevant
articles and read more top articles as suggested by the system. The phenomenon of relying on the
recommendations is named algorithm dependence, similar to media dependency [27], where users are
dependent on media to select news information. People with algorithm dependence are more likely to
consume the news articles suggested by recommender systems, including consuming more top articles.
Thus, we propose:
H6: Information seeking is correlated with higher goal commitment compared with
information scanning.
H7: Higher goal commitment is correlated with a) fewer irrelevant articles read and b) more
top articles read.
H8: Higher goal commitment is directly correlated with a) lower information overload and b)
higher user satisfaction.
H9 Higher goal commitment is indirectly correlated with a) lower information overload and b)
higher user satisfaction.
3. Method
3.1 Experimental Design, Manipulation, and Experimental Materials
The study contained a 2 (information behavior: seek/scan) X 2 (relevance: high / low) experimental
design to test the impacts on information overload and user satisfaction. Participants in the information
scanning group scanned twenty articles while participants in the information seeking group were
required to type keywords in a search bar and search for articles (see Figure 1). Participants were
randomly assigned to seek/scan articles on one of the two topics, which were “Why am I so tired?” and
“What causes high blood pressure?” The two topics were both listed in the top 10 health questions
people searched on Google in 2017 [28] and 2018. No certain causes were affirmed by Google while
multiple causes of tiredness and several risk factors of high blood pressure were introduced in the
articles. 10 news irrelevant to either topics were selected from the health section of Google news.
Articles were between 500-600 words long.
Figure 1: First page of the web portal from the scanning condition.
Participants in high relevance group were exposed to sixteen articles relevant to their topic and four
articles irrelevant to their topic, among them the top three articles were all relevant articles.
Correspondingly, the low relevance group were exposed to ten relevant articles and ten irrelevant
articles, where two of the top three articles were irrelevant articles. All the articles were randomly
selected and sorted despite manipulating the number of relevant articles presented on the page. In other
words, the manipulation controlled the relevance of articles in the top three positions rather than
specifying a position for a certain article. For instance, the top three articles were relevant articles in
high relevance condition. The three relevant articles were randomly selected from the dataset and the
top three articles were randomly sorted. The rest of the articles were also randomly selected from the
dataset and randomly sorted.
Regardless of the experimental conditions, all participants were required to choose 3 articles from
the news portal to read. They were given instructions before entering the portal (See Figure 2). Once in
the portal, participants could see news article headlines with a small preview of the article and were
required to click “Read More” if they wanted to select the article to read (See Figure 3). Three pop-ups
were designed to remind participants about their reading status. For instance, a popup after participants
selected the second article that said, “You just selected the second article. Please read the article then
select 1 more article most relevant to the causes of being tired” (See Figure 4). A button “click here to
go to survey” appeared at the bottom of the third article selected by the participants. Participants were
then sent to a questionnaire in Qualtrics. Users answered a manipulation check question and an attention
check question then proceeded to answer the survey measures detailed below.
Figure 2: Second page of the web portal.
Figure 3: General view of the news page.
Figure 4: Second pop-up instructing users to read one more article.
Eight web portals were designed to conduct the experiment. The information seeking group were
directed to type in keywords in a search bar to proceed to the portal. All the participants were required
to read three articles. Afterwards, they proceeded to the questionnaire in Qualtrics. The content was set
to be 70% width on desktop browsers and full width in mobile browsers. Web log data was inadvertently
collected by the experimental news portal software, including the condition, the sequence of articles
selected, the specific articles selected, time reading each article, and search term for those in the seeking
condition.
3.1.1. Experimental Materials
The materials were gathered from an actual Google search using the queries “Why am I so tired?”
and “What causes high blood pressure?” separately. All the articles returned on the first two pages were
collected on June 24, 2019 and July 4, 2019 and tailored. In total, nineteen articles about being tired
and sixteen articles about high blood pressure were collected. Ten articles irrelevant to either topic were
also collected on June 24, 2019 and July 4, 2019 from Google News under the category of health news
(sorted by date). Topics that contained information that might elicit strong negative emotions were
excluded. The lengths of all articles were tailored to vary from 510 to 599 words. Sixteen articles
relevant to each topic and the ten irrelevant articles were retained.
3.1.2. Participants
G*Power recommended a minimum sample size of 619 to test the proposed hypothesis. 834
participants were recruited from Amazon MTurk. Survey data with wrong answers to the attention
check question, duplicate IP, and incomplete ones were filtered out. After merging the web log data and
the survey data, incomplete ones, such as the cases that do not contain web log data, were deleted. In
addition, 14 cases from the seeking condition that did not follow the instructions to type in relevant
keywords in the search bar were deleted. The final dataset contains 669 participants. Participants were
18-74 years old. 35.7% of them were males, and 62.3% were females. 44.1% of participants have
college and above education level and 75.8% were white.
3.1.3. Manipulation Check
Participants answered manipulation check questions when they entered the questionnaire after they
completed reading the news articles in the experimental news portal. Participants answered three
questions regarding the scanning and seeking condition including, “I typed in keywords and searched
for information about [certain topic]”, “The web portal retrieved news about [certain topic] after I typed
in keywords”, and “The web portal presented information about [certain topic] without asking me to
type in keywords”. The scanning score was calculated by the average of the reversed score of the first
question and the third question. The seeking score was calculated by the average of the first two
questions. As expected, the scanning condition showed a significantly higher score (M = 5.09, N =
333) on scanning than seeking group (M = 2.61, N = 336) (F(1, 665) = 334.56, p < .000). For the same
groups, the seeking condition showed a significantly higher score (M = 6.07, N = 336) on seeking than
scanning group (M = 3.37, N = 333) (F(1, 665) = 394.96, p < .000).
Participants were also asked two manipulation check questions regarding the news article relevance
manipulation including, “A majority of articles I saw on the web portals are relevant to the topic of
[certain topic]”, and “The first three articles on the top of the web portal are all relevant to the topic of
[certain topic]”. Answers vary from strongly disagree to strongly agree. The score of relevance is the
average of the two questions. Participants in the high relevance condition (conditions 1-4) showed a
higher score (M = 5.99, N = 262) than low relevance condition (M = 4.03, N = 272). The difference
between groups is significant (F(1, 665) = 299.54, p < .000).
4. Measures
4.1 Information Overload
Information overload (M = 2.80, SD = 1.22) was measured using a four-item measure adapted from
Stephens and Rains [29]. Participants rated statements indicating that the information they received
through seeking or scanning needed too much explanation to be useful, the information they received
was too much information to process, the information they received was more information than they
needed, and the information they received was about the right amount of information participants
needed. Information overload was reported in a seven-point Likert-scale ranging from 1 = strongly
disagree to 7 = strongly agree. The last item was reversed coded. The scale was reliable ( = .78).
4.2 Goal Commitment
Goal commitment (M = 5.13, SD = .88) was measured by the scale created by Hollenbeck, Williams
and Klein [30]. The scale contained nine items, asking participants their feelings of finding information
in the news portal. Sample Items included “It is hard to find news information”, and “It was unrealistic
for me to expect to find news information I’m interested in.”. Several items were reverse coded guided
by the scale instructions. However, one item in the scale was removed due to lack of reliability ( = .55).
After deleting the item “It was quite likely that the process of finding news information may need to be
revised, given how the information process went”, the scale was reliable ( = .72). The final scale used
in the paper had eight items.
4.3 Satisfaction
Satisfaction (M = 5.34, SD =.86) was measured by the Psychological Need Satisfaction and
Frustration Scale [31]. The scale consisted of four subscales measuring satisfaction of autonomy,
frustration of autonomy, satisfaction of competence, and frustration of competence. Each sub-scale
contained four items and the overall scale contained sixteen items. Examples of the items were “I felt a
sense of choice and freedom when I was looking for information on the news portal.” and “I felt that
my decision of selecting certain news information reflected what I really wanted.”. Items in the
subscales of frustration of autonomy and competence were reverse coded ( = .89).
4.4 Top Articles Selected by Users
Top articles selected was calculated using the participants’ digital trace from the web portals. The
web portals recorded the sequence of news articles and the article number selected by the users
sequentially. The top articles refer to the number of the top three randomly displayed articles in the web
portal that were selected by the participant. The number ranged from 0 – 3. Participants read 1.43 top
articles on average (SD = .91).
4.5 Irrelevant Articles selected by Users
Similar to top articles selected by users, irrelevant articles selected by users were calculated based
on the number of irrelevant articles selected by the user from the digital trace data. Results ranged from
0 – 3. Participants read .34 irrelevant articles on average (SD = .68).
5. Data Analysis Plan
The data analysis was completed using OLS regression. The indirect effects in the mediation
models were calculated using the PROCESS macro v3.4 within SPSS Statistics 26 (Hayes, 2018). Four
mediation model analyses were examined by PROCESS (model 4). The first two models started with
the independent variable relevance, mediators a) numbers of irrelevant articles selected and b) numbers
of top articles selected, ended with dependent variables information overload (model 1) and user
satisfaction (model 2). The other two models started with mediator information seeking/scanning that
impact goal commitment, independent variable goal commitment, two mediators a) numbers of
irrelevant articles selected, and b) numbers of top articles selected, ended with dependent variable
information overload (model 3) and user satisfaction (model 4).
6. Results
Four models were tested. The results and statistics of the models are reported in figures 5-8. The
first two models tested the relationship between relevance of recommendations, the number of
irrelevant articles read, the number of top articles read and information overload (see Figure 5) and user
satisfaction (see Figure 6). The results showed a negative relationship between relevant
recommendations and the number of irrelevant articles (H1a) and a positive relationship between
relevant recommendations and the number of top articles read (H1b). H1 was supported. Irrelevant
articles read by users was positively related with information overload (H2a) and negatively related
with user satisfaction (H2b). H2 was supported. More top articles read by users was not related with
information overload (H3a) or user satisfaction (H3b) as predicted. H3 was not supported. Relevance
was positively related with information overload (H4a), while is not related to user satisfaction (H4b).
H4 received partial support. High relevance of recommendations indirectly correlated with lower
information overload (H5a) and higher user satisfaction (H5b). H5 was supported.
Figure 5: Model 1: How relevance impacts information overload.
Note: * p <.05; ** p < .01; *** p < .001
Figure 6: Model 2: How relevance impacts user satisfaction.
Note: * p <.05; ** p < .01; *** p < .001
Models with seeking/scanning as the independent variable related to goal commitment, the number
of irrelevant articles read, the number of top articles read and information overload (see Figure 7) and
user satisfaction (see Figure 8) were also tested. The results did not indicate a relationship between
seeking and goal commitment (H6). H6 was not supported. Goal commitment was negatively correlated
with irrelevant articles (H7a) and positively correlated with top articles read (H7b). H7 was supported.
Goal commitment is directly negatively correlated with information overload (H8a) and positively
correlated with user satisfaction (H8b). H8 was supported. Goal commitment was also indirectly
negatively correlated with information overload (H9a) and positively correlated with user satisfaction
(H9b). H9 was supported.
Figure 7: Model 3: How seeking/scanning impacts information overload.
Note: * p <.05; ** p < .01; *** p < .001
Figure 8: Model 4: How seeking/scanning impacts user satisfaction.
Note: * p <.05; ** p < .01; *** p < .001
7. Discussion
The paper explored the impacts of news recommendation relevance, information scanning vs.
seeking, and goal commitment on information overload and user satisfaction via news selection.
Existing research has rarely indicated how news selection processes are impacted by both algorithmic
settings and personal factors, such as individual information behaviors or goal commitment, leaving a
blank in sequential recommender systems research. The research also provides findings on user
behaviors the context of in search engine recommendations.
Overall, high news recommendation relevance and high goal commitment predicts low information
overload and high user satisfaction via news selection. Reading behaviors including scanning and
seeking did not impact goal commitment, contrary to our expectation. However, goal commitment
positively impacted news selection, information overload, and user satisfaction. In addition, the design
of the recommender systems, including the relevance for the top three articles and overall relevance in
the recommendations impacted users’ news selection. Users’ news selection predicted information
overload and partially user satisfaction. These results indicate that the design of recommender systems
should consider focusing on user motivations, such as goal commitment.
The first finding was regarding news recommendation relevance, news selection, information
overload, and user satisfaction. The results indicated that users read fewer irrelevant articles when being
exposed to a great proportion of highly relevant recommendations. The finding is consistent with the
existing literature that has found that recommendations influence users’ news selection where users are
more likely to read popular stories in the recommendation condition [32]. However, the finding also
illustrated that user selection might be limited by recommended content since users can only select from
the recommendations. We expand the existing literature, by illustrating the significant impacts of
relevant recommendations on users’ news selection and subsequent information overload and
satisfaction perceptions. What’s more, the finding indicated that when being exposed to more irrelevant
articles, users selected more irrelevant articles. This, in turn, led to greater information overload
perceptions. The reason for this finding might be choice overload, where users confronting similar
related-articles recommendations and experience difficulties selecting certain articles for use [9].
Another explanation might be the reliance on news recommendations. The peripheral processing route
in the dual process theories [33] helps explain the mechanism when users limits cognitive effort to make
a decision on news selection. Users are more likely to rely on recommendations when making news
selection to minimize cognitive load.
Surprisingly, users did not incline to select more top articles when being exposed to high relevant
recommendations. Although existing research found the significant impacts of ranking on user’s news
selection [34], and the merits of top N recommendations on identifying items fit users’ tastes [35], the
current study found that users might not always select top articles in more relevant recommendation
conditions. Users actively select news articles with more relevant recommendations, which can be
explained by the central route of dual processing model, where users are aware of and deliberately make
choices. That is, when news recommendations are more relevant, users may have greater cognitive
resources to peruse the full list of articles. More relevant recommendations might be a key factor to
more freedom in news selection. Trust towards recommendations may cause more top articles selection
as well. More trust on the recommendations led to more top articles selection. Cognitive load might be
another cause of more top articles selection. Users with high cognitive load were more likely to select
more top articles to save energies. We recommend future research explore trust as another potential
explanatory variable.
Second, users’ news selection in high or low relevance conditions impacted perceptions of
information overload and satisfaction. Users are more likely to be more information overloaded, and
less satisfied with their information autonomy when they read more irrelevant articles and more top
articles. Consistent with existing research [3], the study also found the correlation between news
recommender systems use and perceived information overload. Users might show self-doubt of their
ability of online seeking/scanning, which caused higher information overload and less user satisfaction.
Users in the low relevance conditions might have experienced frustration towards the recommendations,
which also might have led to higher information overload and less user satisfaction.
In addition, the study also showed that users’ news selection was correlated with information
overload and user satisfaction, providing a more detailed description between news recommender
systems and user interaction as proposed by [36]. Contradicting our prediction, reading more top articles
lead to less user satisfaction. User autonomy of news selection might explain it, where users are unlikely
to continue their interests in exploring the same topic with top N recommendations design.
Third, indirect correlations between relevance of recommendations and information overload were
found. These indirect effects indicated that both recommendations and user’s news selection impacted
user experience in recommender systems. To design a personalized sequential recommender system,
both algorithmic rules and users’ news selection choices, which is also called the implicit user feedback,
should be considered.
Browsing style (seeking or scanning) was proposed to indirectly impact information overload and
user satisfaction via goal commitment. These relationships were not found in the study’s results. In
other words, users seeking or scanning news recommendations in search engines or web portals did not
show different levels of information overload and user satisfaction. This finding might indicate that
users spend similar amounts of effort looking for information when seeking or scanning for certain
topics. Different recommender rules for seeking or scanning behaviors might not be a necessary.
Interestingly, the study found that users with higher goal commitment is correlated with fewer
irrelevant articles and fewer top articles read. Uses and gratifications theory helps explain the news
selection that users with high goal commitment were more likely to meet their information requirement
when reading fewer irrelevant and top articles. Users might enjoy the exploration process of finding
relevant information in either high or low recommendation conditions. It also indicates that
recommender systems design needs to consider the level of users’ goal commitment to provide a more
personalized information retrieval atmosphere either with more top N recommendations or not.
Consequently, users with high goal commitment are more likely to be less information overload and
high user satisfaction. These results support the notion that users exerting control over
recommendations [37] help relieve information overload.
The paper explored how users’ news selection were impacted by both algorithmic rules and goal
commitment, and impacted information overload and user satisfaction. It provided empirical evidence
regarding how relevant impacts information overload and user satisfaction. What’s more, it illustrated
the detailed news selection process that increased the understanding about the interactions between
users and recommendations. The study provided practical guidance to users that a higher goal
commitment and agency leads to a more satisfied online information experience. We recommend
system designers find ways to motivate their users’ to be more deliberative while using their online
recommendation systems. The study also contributed to the design of recommender systems, where
users’ news selection and personal factor, such as goal commitment could be considered as potential
input of personalized algorithms. In addition, top recommendations do not always lead to higher user
satisfaction. The study also found the potential reliance on recommendations, which calls for more user
behavior studies.
Like all the other studies, the study owns its limitations, and we hope to conduct future research to
fill in the gaps. First, we didn’t manipulate goal commitment, while we did find it to be a key variable
driving satisfaction and information overload. Future studies should focus on how to manipulate goal
commitment. Second, we only consider news recommendation relevance. Future research could
examine diverse conceptualizations of relevance and its impacts. Third, we only consider content
filtering. Future studies could look at collaborative filtering and other filtering rules. Forth, future
studies could explore when seeking and scanning differ in the outcomes of information behaviors. Fifth,
future studies could also include more news selection measures and explore how user react to
homogenous content and cognitive load. The study could also be further expanded to include sequential
reading pattern where users’ reading pattern is linked to information overload and user satisfaction. The
exploration could provide a direct picture of first-hand user interactions towards the news
recommendations. The study could also be replicated in other contexts, such as political news
recommendations, for generalization.
8. Conclusion
In summary, the study explored how relevance of news recommendations and information scanning
or seeking impact users’ news selection, and information overload and user satisfaction. The study
found the potential user reliance on recommendations and vulnerability of users receiving low quality
recommendations. Recommendation relevance is a key factor impacting information overload and user
satisfaction in the recommendation system. High relevance leads to lower information overload. In
addition, user control is one factor to consider when designing a recommender system. Lastly,
individual differences, such as differing goal commitment during the information selection process, is
worth considering in recommendation system designs.
9. Acknowledgements
The research was funded by the John Paul Jones Memorial Trust Award in the Department of
Communication at the University of Wisconsin-Milwaukee.
10.References
[1] J. Freyne, S. Berkovsky, E. M., Daly, and W. Geyer, Social networking feeds: Recommending
items of interest, in: Proceedings of the Fourth ACM Conference on Recommender Systems, 2010,
pp. 277-280.
[2] K. Koroleva, A. J. B. Röhler, Reducing information overload: design and evaluation of filtering &
ranking algorithms for social networking sites, in: ECIS 2012 Proceeding, 2012, pp.12.
[3] Jörg Matthes, Kathrin Karsay, Desirée Schmuck, and Anja Stevic. “ ‘Too much to handle’: Impact
of mobile social networking sites on information overload, depressive symptoms, and well-being.”
Computers in Human Behavior 105 (2020). https://doi.org/10.1016/j.chb.2019.106217
[4] Ramondt, Steven, and A Susana Ramírez. “Assessing the impact of the public nutrition
information environment: Adapting the cancer information overload scale to measure diet
information overload.” Patient education and counseling vol. 102,1 (2019): 37-42.
doi:10.1016/j.pec.2018.07.020
[5] R. Cohen, N. Sardana, K. Rahim, D. Y. Lam, M. Li, and O. Maccarthy, Reducing information
overload in social networks through streamlined presentation: a study of content-centric and
person centric contexts towards a generalized algorithm, in: 3rd Workshop on Incentives and
Trust in E Communities, 2014, pp.13-18.
[6] M. Chen, Y. Chen and C. Tseng, "SocFeedViewer: A Novel Visualization Technique for Social
News Feeds Summarization on Social Network Services," in 2013 IEEE 20th International
Conference on Web Services, Honolulu, HI, USA USA, 2012 pp. 616-617.
doi: 10.1109/ICWS.2012.76
[7] Ae Ri Lee, Soo-Min Son, Kyung Kyu Kim. Information and communication technology overload
and social networking service fatigue: A stress perspective. Computers in Human Behavior,vol.
55(A) (2016): 51-61.https://doi.org/10.1016/j.chb.2015.08.011
[8] Beam, Michael A., and Gerald M. Kosicki. “Personalized News Portals: Filtering Systems and
Increased News Exposure.” Journalism & Mass Communication Quarterly, vol. 91, no. 1, Mar.
2014, pp. 59–77, doi:10.1177/1077699013514411.
[9] Felix Beierle, Akiko Aizawa, and Joeran Beel. Exploring Choice Overload in Related-Article
Recommendations in Digital Libraries. 2017. ArXiv:1704.00393 [Cs].
http://arxiv.org/abs/1704.00393
[10] Bawden, David, and Lyn Robinson. “The Dark Side of Information: Overload, Anxiety and Other
Paradoxes and Pathologies.” Journal of Information Science, vol. 35, no. 2, Apr. 2009, pp. 180–
191, doi:10.1177/0165551508095781.
[11] H. Butcher, In meeting managers+ information needs. 1998. pp. 53. London: CIMA.
[12] Angela Edmunds, and Anne Morris. “The problem of information overload in business
organizations: a review of the literature.” International Journal of Information Management, vol.
20, no.1, 2000, pp.17–28, doi: https://doi.org/10.1016/S0268-4012(99)00051-1.
[13] Elizabeth J. Burge. Learning in computer conference contexts: The learners’ perspective. Journal
of Distance Education, vol. 9, 1994, pp.19-43.
[14] Joel Rudd, and Mary Jo Rudd. “Coping with information load: User strategies and implication for
librarians”. College and Research Libraries, vol. 47, no.4, 1986, pp. 315-322.
[15] Saima Khalid, Mihammad Saeed, and Sehrish Syed. “Impact of information overload on students’
learning: An empirical approach.” FWU Journal of Social Sciences, vol. 10, no.1, 2016, pp. 58–66.
[16] O. E. Klapp, Overload and boredom: Essays on the quality of life in the information society.
Connecticut: Greenwood Press. 1986.
[17] S. R. Hiltz, and M. Turoff. “Structuring computer-mediated communication systems to avoid
information overload.” Communications of the ACM, vol. 28, no.7, 1985, pp. 680-689.
[18] S. K. Lee, K. S. Kim, and J. Koh. “Antecedents of news consumers’ perceived information
overload and news consumption pattern in the USA.” International Journal of Contents, vol. 12,
no.3, 2016, pp. 1-11.
[19] C. E. Beaudoin. “Explaining the relationship between Internet use and interpersonal trust: Taking
into account motivation and information overload.” Journal of Computer-Mediated
Communication, vol. 13, no. 3, 2008, pp. 550-568
[20] M. Pazzani, M. uramatsu, and D. Billsus, Syskill & webert: Identifying interesting web sites. In:
Proceedings of the 13thNational Conference on Artificial Intelligence, 1, 1996, pp. 54-61.
[21] F. Ricci, L. Rokach, B. Shapira, and P. B. Kantor, Recommender Systems Handbook. Springer
US, 2011.
[22] Z. Zolaktaf, R. Babanezhad, and R. Pottinger, A generic top-N recommendation framework for
trading-off accuracy, novelty, and coverage, in:2018 IEEE 34th International Conference on Data
Engineering (ICDE), 2018, pp. 149–160. doi: https://doi.org/10.1109/ICDE.2018.00023.
[23] L. Coba, P. Symeonidis, & M. Zanker, Replicating and improving top-N recommendations in
open source packages, in Proceedings of the 8th International Conference on Web Intelligence,
Mining and Semantics, 2018, pp.1–7. doi:https://doi.org/10.1145/3227609.3227671.
[24] D. O. Case, Looking for information: A survey of research on information seeking, needs, and
behavior (2nd edition), UK: Elsevier Ltd, 2007.
[25] O. D. Bruijn, and R. Spence, “A new framework for theory-based interaction design applied to
serendipitous information retrieval.” ACM Transactions on Computer-Human Interaction, vol. 15,
no.5, 2008, pp. 1-38.
[26] E. Katz , J. Blumler, and M. Gurevitch. “Uses and gratifications research.” The Public Opinion
Quarterly, vol. 37, no.4, 1973, pp 509-523.
[27] S. J. Ball-Rokeach, and M. L. DeFleur. “A Dependency Model of Mass-Media Effects.”
Communication Research, vol. 3, no. 1, Jan. 1976, pp. 3–21,
doi:10.1177/009365027600300101.
[28] R. H. Shmerling, Dr. Google: The top 10 health searches in 2017. Harvard Health Blog. URL:
https://www.health.harvard.edu/blog/google-top-10-health-searches-2017-2018022113300
[29] Stephens, Keri K., and Stephen A. Rains. “Information and Communication Technology
Sequences and Message Repetition in Interpersonal Interaction.” Communication Research, vol.
38, no. 1, Feb. 2011, pp. 101–122, doi:10.1177/0093650210362679.
[30]John Hollenbeck, Charles Williams, and Howard Klein. (1989). "An Empirical Examination of the
Antecedents of Commitment to Difficult Goals." Journal of Applied Psychology. 74.(1989): 18-23.
10.1037/0021-9010.74.1.18.
[31] Beiwen Chen, Maarten Vansteenkiste, Wim Beyers, Liesbet Boone, Edward L. Deci, Jolene Van
der Kaap-Deeder, Bart Duriez, Willy Lens, Lennia Matos, Athanasios Mouratidis, Richard M.
Ryan, Kennon M. Shldon, Bart Soenens, Stijn Van Petegem, and Joke Verstuyf. 'Basic
psychological need satisfaction, need frustration, and need strengh across four cultures."
Motivation and emotion 39 (2015): 216-236.
[32] JungAe, Yang. (2016). "Effects of Popularity-Based News Recommendations (“Most-Viewed”)
on Users’ Exposure to Online News." Media Psychology, 19.2 (2016): 243–271.
https://doi.org/10.1080/15213269.2015.1006333
[33] R. E. Petty, and J. T.Cacioppo, (1986). The elaboration likelihood model of persuasion, in: L.
Berkowitz (Ed.), Advances in Experimental Social Psychology. Academic Press, 1986, vol. 19, pp.
123–205. https://doi.org/10.1016/S0065-2601(08)60214-2
[34] Julian Unkel, and Alexander Haas, "The effects of credibility cues on the selection of search engine
results." Journal of the Association for Information Science and Technology, vol.68, no.8, 2017,
pp. 1850–1862. https://doi.org/10.1002/asi.23820
[35] Chen, Y. (n.d.). Learning for top-N recommendations. Retrieved August 5, 2021, from
https://dare.uva.nl/personal/pure/en/publications/learning-for-topn-recommendations(16d7f634-
6cdf-4d45-a6d4-523be2098930).html
[36] Donghee Shin, "How do users interact with algorithm recommender systems? The interaction of
users, algorithms, and performance." Computers in Human Behavior, Vol. 109, 2020.
[37] Sarah Dean, Sarah Rich, and Benjamin Recht. 2020. Recommendations and user agency: the
reachability of collaboratively-filtered information. In Proceedings of the 2020 Conference on
Fairness, Accountability, and Transparency (FAT* '20). Association for Computing Machinery,
New York, NY, USA, 436–445. DOI:https://doi.org/10.1145/3351095.3372866