Emotional Experience during Human-Computer Interaction: A Survey

Abstract As human-computer interaction (HCI) technology becomes more and more integrated into our daily life, increasing attention has been drawn towards the interaction experience in addition to HCI efficiency. In the present study, we conducted a survey to explore context-specific emotional experience in HCI. Four hundred participants were recruited to report the frequency of their emotional experiences on 44 fine-grained emotion items in six representative HCI scenarios. Compared with six matched human-human interaction (HHI) scenarios used as control, the HCI scenarios were in general more frequently associated with negative emotions, and less frequently associated with positive emotions, especially when computer served as a tool for communication with other people. Furthermore, the 44 emotional experience items in HCI were summarized as five factors, representing low-arousal focused, positively engaged, emotionally empathetic, high-arousal negative and frustratingly confused. Our study presents a comprehensive overview of context-specific emotional experience in human-computer interactions and provides a framework for emotion evaluation in HCI applications.


Introduction
Human-computer interaction (HCI) technology has been increasingly involved in our daily life.While previous HCI research has primarily focused on improving interaction efficiency (Brave & Nass, 2007;Picard, 2010), the advances in automation and intelligence of HCI technology have led to more and more research attention paid to users' subjective experience, especially emotional experience in HCIs (Jerritta et al., 2011;Picard, 2003), which has great value for the evaluation and optimization of HCI systems.Most existing research on emotion in human-computer interactions adopted classical unified emotion theories as the framework (Fragopanagos & Taylor, 2005;Ramakrishnan & El Emary, 2013;Sreeja & Mahalakshmi, 2017).The six-basic-emotion theory (i.e., happiness, sadness, fear, anger, surprise, and disgust; Ekman & Friesen, 1978) or its variations (Khalil et al., 2019) has been the most widely used framework for studying emotional experience in a number of typical HCI scenarios, including video games (Piana et al., 2014), online learning (Hasnine et al., 2021), online community engagement (Colneri� c & Dem� sar, 2018), interaction with robots (Devillers et al., 2015), and virtual reality environments (Hinkle et al., 2019).
While unified emotion frameworks have played a dominant role in emotion studies, the context-dependent nature of emotions is gaining increasing interest in recent years.According to the Conceptual Act Theory, emotions are not unified responses, but rather are flexibly constructed by the individual's conceptualization of the situation, as well as past experiences and cultural background.This theory emphasizes the context-dependent nature of emotions and the active role that individuals play in constructing their emotional experience (Barrett, 2014;Hu et al., 2022).In line with this theory, a series of influential survey studies by Cowen and colleagues have investigated the multiplicity of emotional experience in different contexts.They employed different emotion induction methods like watching videos (Cowen & Keltner, 2017) and listening to music (Cowen & Keltner, 2020).Thirty-four and 28 emotion items were included in these studies to provide a comprehensive description of possible emotional experiences associated with these contexts.In each study, they asked subjects to rate each material on multiple emotional dimensions and then clustered the emotions in a semantic space.Their findings revealed that there were considerable differences in emotional experiences depending on the context.Specifically, emotions, like admiration, empathetic pain, amusement, awe, and embarrassment, obtained from watching video scenes, did not appear during listening to music, while emotions, like dreamy, energetic, and annoying, did not appear during watching videos.These studies provided solid evidence for context-specific emotional experiences in a wide range of emotional events.In another influential research in learning science, Pekrun and Linnenbrink-Garcia (2012) conducted an in-depth survey to students' emotional experiences while they were attending classes, studying, and taking exams in academic contexts through qualitative interviews and questionnaires.They identified nine specific categories of "academic emotions," including enjoyment, hope, pride, relief, anger, anxiety, shame, hopelessness, and boredom.Jointly, these lines of studies provided strong evidence to the Conceptual Act Theory, that emotional experience is complex and deeply grounded in situations or contexts.Therefore, when it comes to human-computer interaction contexts, the six-basic-emotion framework may not be sufficient for depicting HCI-specific emotional experiences, but it is necessary to take more emotion items and various HCI scenarios into consideration.
Based on the Conceptual Act Theory (Barrett, 2014;Barrett et al., 2015), we can speculate that emotions arising from the conceptualization of HCI scenarios may be different from those in daily interpersonal activities, because the object of interaction is a computer rather than a real human in HCI scenarios.Previous research has demonstrated that people act differently when they interact with a real person as compared with when they interact with a computer (Liu et al., 2023;Shechtman & Horowitz, 2003).And this difference in behavioral orientations may affect individual's emotional experiences (Camodeca & Nava, 2022;Jones, McGarrah & Kahn, 2019;Lang, 1984).Take online learning as an example, students reported higher frequencies of experiencing emotions, like concentration, flow, boredom and frustration (Baker et al., 2010;D'Mello, 2013), which constituted a pattern distinct from live learning with real people.Moreover, Walter et al. (2014) identified 20 emotion items to be highly relevant for HCI scenarios.However, these studies have focused on a single or relatively restricted HCI contexts, and a comprehensive depiction of HCI-specific emotional experience still requires a deeper look into the complexity and diversity of HCI contexts.According to the Activity Theory (Bedny & Karwowski, 2011;Kuutti, 1996;Leont'ev, 1978;Uden, 2007), there exist substantial differences among different types of HCI tasks, which can be generally divided into two types, computer as an interaction object (Cowie et al., 2001;Fragopanagos & Taylor, 2005;Hibbeln et al., 2017) and computer as a communication tool (Fabri, Elzouki & Moore, 2007;Wadley et al., 2015).Nevertheless, it still remains unclear whether there exist distinct emotional experiences in these subtypes of HCI contexts.
Moreover, the existing emotion categorization may not be suitable for describing more daily HCI scenarios, especially those with complex emotional experiences.It has been shown that people in real-life situations often experience two or more discrete emotions simultaneously (Ellsworth & Smith, 1988;Hu et al., 2022;Martinent et al., 2012).For instance, people may experience a mixture of joy and sad during their graduation ceremony, but feel both fear and disgust when they see a mouse in the room.This co-occurrence nature of human emotional experience, however, has also been overlooked in existing HCI studies, which mainly focused on single discrete emotions.To effectively summarize the co-occurrence characteristics of complex emotional experience in HCI contexts, it is necessary to employ multivariate data analysis methods such as factor analysis, to explore correlations between multiple emotions.The factorization would reveal co-occurrence relationships of emotions in HCI contexts, thus helping us define more representative HCI emotional states that are independent of each other.
Boosted by the aforementioned motivations, in the present study, we employed a survey method to investigate emotional experience specific to HCI scenarios.To acquire a relatively fine-grained description of emotional experience in HCI contexts, we chose 44 emotion items as the scoring dimensions in the survey based on recent advances in emotion categorization (Baker et al., 2010;Cowen & Keltner, 2017, 2020;Craig et al., 2004;Kort et al., 2001).We also included human-human interaction (HHI) contexts as control conditions to investigate the HCI specificity of the emotional experiences.As interacting with a computer would have less social-related consequences (Liu et al., 2023;Shechtman & Horowitz, 2003), it is expected that HCI would be less strongly linked with social emotions, such as admiration, adoration, etc.Furthermore, considering the diversity of HCI scenarios, we included six representative HCI scenarios, which can be further divided into two major subtypes according to the role computer plays in HCI, i.e., computer as an interaction object and computer as a communication tool (Bedny & Karwowski, 2011;Kuutti, 1996;Leont'ev, 1978;Uden, 2007).This subdivision of HCI contexts enabled us to further elucidate the possible contributions of context-specific effect to the difference in emotional experience between HCI and HHI.As interactions with a computer are less bonded with instant social feedback (Hibbeln et al., 2017;Wadley et al., 2015), we hypothesized that HCI contexts, especially the computer-as-an-interaction-object scenarios (Wu et al., 2023), may give rise to less positive emotional experience in contrast with HHI contexts.In addition, we also employed an exploratory factor analysis to disentangle the co-occurrence relationships of emotions and distill more representative emotional states specific to HCI contexts.

Methodology
This study employed quantitative questionnaires to investigate fine-grained emotional experience in HCI and HHI scenarios.

Participants
The valid data were obtained from a total of 400 participants, with N ¼ 200 participants for questionnaire#1 (125 females, mean age 27.60 ± 5.74) and N ¼ 200 participants for questionnaire#2 (114 females, mean age 27.37 ± 5.74, see "2.2.Procedure" for the explanations of the two questionnaires).Each participant was paid 18 RMB (approximately equal to 3 USD).All participants gave informed consent, and the protocols of this study were approved by the local ethics committee at the Department of Psychology, Tsinghua University.

Procedure
To have a fine-grained description of emotional experience, we included 25 positive and 19 negative emotion items in our questionnaire for users to rate the frequency of each emotion experienced in different themes of scenarios.Participants were required to answer how frequently they experienced each emotion in 3 HCI scenarios and 3 matched HHI scenarios.To reduce the number of questions for each participant, the participants were randomly assigned to questionnaire#1 or questionnaire#2.For questionnaire#1, there were online and offline shopping, banking, and chatting scenarios; and for questionnaire#2, there were online and offline learning, conference, and navigation scenarios.It should be noted that although there were 264 questions (6 scenarios (3 HCI þ 3 HHI) � 44 emotion items ¼ 264) per questionnaire, the participants only need to read 6 scenario information to complete 44 questions per scenario.The time required to complete each questionnaire was about 6 to 10 minutes.
The questionnaires were created and published on an online data collection platform Credamo.When publishing the questionnaires, we set a data collection target of 200 valid samples for each questionnaire.To ensure the quality of the completed questionnaires, we limited participants to those who had more than 80% qualified rate of their previous responses on Credamo.Responses were checked for each participant after data collection.We excluded samples with significantly low-quality responses (based on the actual questionnaire completion performance, e.g., selecting the same options for all questions or taking less than 5 minutes to complete the questionnaire) and republished questionnaires until the target sample size of valid data was achieved.A total of 436 participants were recruited for the questionnaires, of which 36 responses were excluded for further analysis because of their low-quality completion.

Questionnaire design
The questionnaires were designed on the basis of discrete emotion theory (Ekman, 1992), instructing participants to rate their experience on each discrete emotion item (P� erez-Espinosa et al., 2022).To include a broad range of emotions, we referred to the emotion items under different induced conditions (Cowen & Keltner, 2017, 2020), and added eight more emotions based on previous HCI emotion studies (see Table 1).
The representative HCI scenarios in the questionnaires were mainly decided on the basis of the analysis report released by Sensor Tower, a mobile applications data analytics company.Their report "Global Mobile Application (Non-Game) Market Outlook 2023" (Sensor Tower, 2023) summarized five representative categories of global nongame applications in 2022, including finance (e.g., online banking), shopping, entertainment, social and productivity (e.g., online learning and online conference) tools.Another two scenarios were included as well, based on a report The State of Travel Apps in 2022 (Sensor Tower, 2022), in which travel and navigation applications showed a surge in global demand.In summary, we chose shopping, banking, chatting, learning, conferencing, and navigation as the six representative HCI scenarios in our study.
To further provide a comparison between HCI and HHI scenarios, we also included six matched HHI scenarios in the questionnaire (see Table 2), corresponding to the six representative scenarios of HCI in terms of event themes.Moreover, we also divided the six HCI scenarios into two types: computer as an interactive object (including shopping, banking, and navigation) and computer as a communication tool (including chatting, learning, and conference), according to the Activity Theory (Bedny & Karwowski, 2011;Kuutti, 1996;Leont'ev, 1978;Uden, 2007).This division is expected to provide further insight about possible difference in emotion experience for these two types of interactions.
Specifically, the questionnaires contained prompts which briefly described the scenarios (consistent with the wording in Table 2) and asked participants to rate the frequency of experiencing each of the 44 emotions on a scale of 1 to 5. Therefore, each participant scored 44 questions for 3 HCI scenarios and 3 HHI scenarios, with a total of 264 questions.For example, participants who were assigned to complete questionnaire #1 needed to answer the question: "How often do you experience the following emotions when you go shopping online?Please rate each emotion on a scale of 1 to 5." The full-length questionnaires were translated into English and provided as Appendix.

Data analysis
Before analyzing the data, we merged the previously obtained data from questionnaire#1 and questionnaire#2 to obtain 200 valid emotional experience data of all scenarios.To verify the reliability of the questionnaire, the parallel-forms reliability was calculated based on the correlation between the responses of questionnaire#1 and questionnaire#2.
To characterize the emotional experience of the HCI scenarios, we performed descriptive statistical analyses-the mean frequency score for each emotion in each scenario was calculated for each participant.In order to understand which emotional experiences in HCI were significantly distinctive from other scenarios, we conducted a paired t-test between HCI and the HHI for each emotion.To correct for multiple comparisons, we used the false discovery rate (FDR) method, and all the reported p-values were FDR corrected (p < 0.05).
To obtain the representative emotional factors of HCI scenarios, exploratory factor analysis was used to derive irrelevant emotional factors.Before the analysis, we conducted a structural validity analysis on the questionnaire data.The number of retained factors was determined by parallel analysis (Franklin et al., 1995;Hayton et al., 2004).Specifically, the eigenvalues of the real data matrix were compared to the eigenvalues of a same-sized Monte-Carlo simulated matrix created from random data to evaluate their statistical significance, and only the factors with significant eigenvalues were retained.In addition, we calculated the correlation matrix between the HHI and the HCI emotion factors based on the loadings of the emotion factors on different emotions, to explore the specificities of the HCIrelated emotion factors.

Context-specific emotional experiences in HCI scenarios
First of all, we checked the reliability of the questionnaire in our study.The parallel-forms reliability of questionnaire#1 and questionnaire#2 was 0.85, indicating an acceptable reliability of questionnaires.The frequency scores of emotional experiences in HCI scenarios and their comparisons with the HHI counterparts are shown in Figure 1.The most frequently experienced positive emotions in both HHI and HCI were concentration and joy, and the most frequently experienced negative emotions were boredom and confusion.Most positive emotions had higher frequency scores in HHI than in HCI scenarios, while more negative emotions had higher frequency scores in HCI scenarios (paired sample t-test, FDR adjusted ps < 0.05).The following positive emotions differed significantly in the frequency among HCI and HHI scenarios: admiration, aesthetic appreciation, adoration, excitement, realization, nostalgia, sympathy, awe, interest, romance, joy, satisfaction, concentration, flow, hopeful, surprise, anticipatory, amusement.Significant differences existed in the following negative emotions: awkwardness, confusion, fear, envy, contempt, disappointment, disgust, indifference, anger, ennui, boredom.

HCI-specific emotion factors
Figure 2(a and b) show the correlation matrix between emotions in HCI and HHI scenarios, where within-valence emotions were more significantly correlated than cross-valence emotions.Strongest correlations were found between emotion pairs of adoration and aesthetic appreciation, ennui and puzzlement.These correlational results provided a solid basis for the planned factor analysis.Moreover, our data contained sufficient shared variance for factor analysis, for both HCI data (KMO ¼ 0.95; Bartlett's test of sphericity, v2 ¼ 22349.72,p < 0.001) and HHI data (KMO ¼ 0.94; Bartlett's test of sphericity, v2 ¼ 21937.99,p < 0.001), indicating a satisfying structural validity.
According to the results of parallel analysis, 3 components of positive emotions and 2 components of negative emotions were retained in the HCI scenarios, while 3 components of positive emotions and 3 components of negative emotions were retained in the HHI scenarios.As shown in Figure 2, the first positive emotion factor of HCI scenarios  was similar to the second factor of HHI.And higher loadings on emotions included concentration, confident, calmness, satisfaction, triumph, and realization (all a > 0.40), which were related to a low arousal focused state.The second positive emotion factor of HCI scenarios was similar to the first factor of HHI.And higher loadings on emotions included joy, amusement, interest, excitement, aesthetic appreciation, and entrancement (all a > 0.50), which were associated with a state of positive engagement.In addition, the third positive emotion factor for HCI had higher loadings emotions including empathy, relief, romance, and nostalgia (all a > 0.50), which were associated with a state of emotional empathy.By contrast, the third positive emotion factor of HHI scenarios loaded higher on emotions including anticipatory, confident and hopeful (all a > 0.50), indicating a state of positively anticipated.
Regarding the results of factor analysis of negative emotions, the first negative factor of HCI was similar to the third negative factor of HHI, with high loadings on high arousal negative emotions including horror, fear, intrigue, guilt and envy (all a > 0.50).The second negative factor of HCI was similar to the first negative factor of HHI, with higher loadings on emotions including confusion, frustration, puzzlement, and anxiety (all a > 0.50), indicating a state of frustratingly confused.Additionally, the second negative factor unique to HHI loaded higher on emotions including indifference, ennui, disgust, and boredom, indicating a state of boredom.

Emotional experiences in HCI scenarios: Computer as a communication tool vs. as an interactive object
To further identify the specific emotional experience in different themes of HCI scenarios, we classified the six HCI scenarios into two types-computer-as-an-interactive-object scenarios (i.e., shopping, banking, and navigation) and computer-as-a-communication-tool scenarios (i.e., online chatting, learning, and conference).We first calculated the difference in frequency score between HCI and HHI conditions for each scenario type and each emotion valence.Therefore, a frequency differential less than zero means this emotion was less frequently experienced in HCI than in HHI scenarios.As shown in Figure 3, in comparison with computer-as-an-interactive-object scenarios, most positive emotions were experienced less frequently, while some negative emotions were experienced more frequently in computer-as-a-communication-tool HCI scenarios.Among the positive emotions, the following emotions differed significantly in the differential scores between computer-as-an-interactive-object and computer-as-a-communication-tool scenarios: admiration, aesthetic appreciation, adoration, excitement, realization, nostalgia, awe, interest, romance, joy, satisfaction, concentration, flow, confident, craving, pride, triumph, entrancement, amusement.As for the negative emotions, significant differences were found in the following emotions: awkwardness, confusion, envy, puzzlement, indifference, ennui, boredom.
We also conducted factor analyses for computer-as-aninteractive-object and computer-as-a-communication-tool scenarios, respectively.As shown in Figure 4, three factors of positive emotions and 2 factors of negative emotions were retained under both types of HCI.Specifically, the first positive emotion factor for computer-as-a-communicationtool scenarios was similar to that of computer-as-an-interactive-object scenarios-high loadings on emotions included satisfaction, joy, interest, hopeful, confident, anticipatory, and concentration (all a > 0.50), indicating an emotional state of positively engaged.The second positive emotion factor of computer-as-a-communication-tool scenarios and the third of computer-as-an-interactive-object scenarios were similar-high loadings on emotions included empathy, romance, relief, and nostalgia (all a > 0.50), indicating a state of emotionally empathetic.In addition, the third positive factor of computer-as-a-communication-tool scenarios had high loadings on the emotions of surprise and awe (all a > 0.50), suggesting a positively surprised state.Lastly, the second positive factor of computer-as-an-interactive-object scenarios had high loadings on concentration and calmness, indicating a positively focused state.
The first negative factor of computer-as-a-communication-tool was a state of bored, loading higher on emotions including boredom, ennui, disgust, frustration, and indifference (all a > 0.50), whereas the first negative factor of computer-as-an-interactive-object is a state of stuckness and incomprehension, with higher loadings on confusion, frustration, puzzlement, anxiety (all a > 0.50), etc.The second negative factor of computer-as-a-communication-tool and computer-as-an-interactive-object scenarios were similar, loading higher on emotions such as horror, contempt, anger, and intrigue (all a > 0.50), inferring a high arousal negative state.

Discussion
Defining the emotions of interest for human-computer interaction (HCI) is challenging due to its context-dependent nature as well as the complexity of HCI scenarios.In the present study, we conducted a survey to explore contextspecific emotional experience in diverse HCI scenarios.Compared with six HHI scenarios used as control, the HCI scenarios were in general more frequently associated with negative emotions including boredom, ennui, and anger, and less frequently associated with positive emotions including admiration, aesthetic appreciation, and adoration.Furthermore, the 44 emotional experience items in HCI were summarized as five factors, representing low-arousal focused, positively engaged, emotionally empathetic, higharousal negative and frustratingly confused.In addition, people experience positive emotions less frequently, but negative emotions more in HCI scenarios, especially when they treat HCI as a tool for communication with other people as compared to treating HCI as an interactive object.
The item-by-item comparison of the emotional experience frequency data between the HCI and the HHI scenarios revealed largely similar but distinct patterns.The similarity between these two types of scenarios was reflected in the high correlations of the overall emotional frequency patterns (Figure 1): top-rated emotions were mostly the same for HCI and HHI.This piece of observation was in line with some previous studies advocating the similarity between HCI and HHI (Shen & Wang, 2022;Siegert et al., 2015;Walter et al., 2014).However, there were also significant differences in quite a number of emotion items, with HCI associated with higher frequencies of negative emotions such as boredom, ennui, and anger, as well as lower frequencies of positive emotions such as admiration, aesthetic appreciation, and adoration.These results could be related to the state-of-the-art development of the HCI technology, which does not always work as well as expected and hereby resulting in more negative emotions such as boredom (Makelberge, 2004;Giakoumis et al., 2010).Alternatively, it is also possible that these results were due to the nature of HCI-computer is considered to be less human-like and therefore less likely to induce those positive social emotions that arise from human interactions, such as admiration (Hareli & Parkinson, 2008).This is consistent with our hypothesis that social emotions could be experienced less in HCI scenarios.More importantly, with relatively diverse HCI scenarios and the relatively complete coverage of emotion items, the present findings are expected to provide a comprehensive overview of the specificity of emotions for HCI contexts.
The comparison between the computer-as-an-interactive-object and the computer-as-a-communication-tool HCI scenarios provide further information regarding how the HCI-specific emotion experience arises.Interestingly, the observed general difference between HCI and HHI seemed to be mainly related to computer-as-a-communication-tool scenarios (Figure 3), which is not in line with our expectation.When computer was considered as a communication tool, the HCI user was expected to be actually communicating with another person.Thus, this result would suggest that the HHI through the state-of-the-art HCI technology (i.e., computer-as-a-communication-tool) is still not comparable to real HHI (Baker et al., 2010;Dunlop & Brewster, 2002;Huang, 2009), leading to less positive and more negative emotions.However, the computer-as-an-interactive-object scenarios showed its unique patterns as well (Figure 3), e.g., less awkwardness and more amusement, but only with a relatively limited number of emotion items.While these results resembled the findings from human-robot interaction studies (Chuah & Yu, 2021;Stock-Homburg, 2022), the relatively limited number of involved emotion items suggested a largely consistent pattern between computer-as-an-interactive-object and HHI scenarios.Taken together, the differences between computer-as-an-interactive-object and computer-as-a-communication-tool HCI scenarios indicate that the observed general differences across all the HCI scenarios might be mainly attributed to the HCI technology per se (i.e., as a communication tool), not by the computer as an interactive object.
More importantly, the factor analysis revealed five HCI-related emotion factors.Among them, low-arousal focused, positively engaged, high-arousal negative and frustratingly confused were shared by HCI and HHI scenarios, suggesting the largely consistent emotional experience in these two types of scenarios (Walter et al., 2014).However, the factor of emotional empathy was relatively distinct for HCI scenarios.A possible explanation is that since HCI takes the form of content interaction through an interactable interface, the lingering and processing time of a single message is much more prolonged, and the emotional content involved is more likely to generate stronger emotional resonance for the user.Previous research in online communities has also mentioned the effect of emotional amplification in HCI scenarios (Goldenberg & Willer, 2023;Luo et al., 2020).There were also two factors specific for HHI: positively anticipated and the boredom factor.According to the previous research (Ludema et al., 1997;Sawyer & Clair, 2022), the anticipated emotion could be defined as an affirmative form of social discourse, indicating the generation of anticipatory is also related to social interaction.And the boredom factor might be a product of social stress (Parasuraman & Purohit, 2000).These findings are consistent with our hypothesis, indicating a more socialrelated emotional experience in HHI scenarios.By further dividing the HCI scenarios into computer-as-an-interactive-object and computer-as-a-communication-tool scenarios, the results were also similar.Positive surprised and bored factors were unique to the computer-as-a-communication-tool scenarios, while low-arousal focused and frustratingly confused factors were unique to the computer-as-an-interactive-object scenarios.The reason may be that when the computer acts as a communication tool, the user interacts more with the content and can be surprised or bored by the information received.Whereas, when the computer serves as an interactive object, the user is task-oriented and interacts more with the technology, thus experiencing more feelings of focused or frustration.In sum, the factor analysis took the co-occurrence nature of human emotion experience into consideration, and the results provide a concise but effective summary of the emotion experience during human-computer interactions.
There are several limitations of the present study that should be noted.First, our concern of emotional experience focused on the frequency of emotions.The inclusion of other characteristics of emotion experience such as intensity would provide further insights into HCI-related emotions (Shui et al., 2021;Zhang et al., 2020), preferably with a careful consideration of a balance between the richness of information and the expected survey time.Second, the HCI scenarios selected in the questionnaires were all from typical mobile phone interactions.Further studies are necessary to include more HCI scenarios such as driving (Braun et al., 2022;Zepf et al., 2021).Third, while the survey method is convenient and flexible, it would be necessary to have laboratory-based experiments with real interaction scenarios (Harrison et al., 2007;Kuutti & Bannon, 2014) to verify the findings in high ecological validity settings.

Conclusion
In the present study, we provide a comprehensive view into the context-specific emotional experience in various types of daily HCI scenarios by taking advantage of a fine-grained survey covering 44 emotion items.In contrast with humanhuman interactions, our findings reveal that some negative emotion items are more specific to HCI scenarios, especially when a computer is used for communicating with other people.Factor analyses further disentangle the co-occurrence characteristics of complex emotional experiences specifically in HCI scenarios.We also elaborate the differences in emotional experience between two subtypes of HCI scenarios, i.e., computer as a communication tool vs. computer as an interactive object.Altogether, this study offers great theoretical value for extending the Conceptual Act Theory of emotion in HCI research, as well as practical value for emotion evaluation in HCI applications.

Disclosure statement
No potential conflict of interest was reported by the author(s).

Figure 1 .
Figure 1.Comparison of mean frequency scores of positive and negative emotions among HCI and HHI scenarios.Emotions are arranged from left to right according to the differences in frequency scores between HHI and HCI scenarios.The red dots indicate the scores in HHI scenarios and the blue dots indicate the scores in HCI scenarios.The bars below indicate the difference between HCI and HHI scores and the color refers to which condition (HCI/HHI) scored higher.Significant results (FDR adjusted p <.05) are starred.

Figure 2 .
Figure 2. Correlations of emotions in HCI (a) and HHI (b) scenarios.The emotions from top to bottom and from left to right are: Anger, Anxious, Awkward, Boredom, Confusion, Contempt, Disappointment, Disgust, Empathic pain, Envy, Fear, Guilt, Horror, Sadness, Puzzlement, Frustration, Ennui, Indifference, Intrigue, Admiration, Adoration, Aesthetic appreciation, Amusement, Awe, Calmness, Craving, Entrancement, Excitement, Interest, Joy, Nostalgia, Pride, Relief, Romance, Satisfaction, Surprise, Sympathy, Triumph, Concentration, Realization, Flow, Hope, Confident, Hopeful.Results of factor analysis of positive (c) and negative (d) emotion frequency scores for HCI scenarios, blue indicates positive correlation, red indicates negative correlation; positive (e) and negative (f) emotion frequency scores for HHI scenarios.Blue indicates a positive loading and red indicates a negative loading.(g) Correlations between the positive emotion factors of HCI and HHI.(h) Correlations between the negative emotion factors in HCI and HHI scenarios.In the correlation plot, blue indicates positive correlation, red indicates negative correlation, and the size of the dots indicates the absolute value magnitude.

Figure 3 .
Figure 3. Differences in frequency score among HCI and HHI scenarios for positive (left) and negative (right) emotions.Significant difference between computer-asan-interactive-object and computer-as-a-communication-tool scenarios are starred (FDR adjusted p <.05).

Figure 4 .
Figure 4. Results of factor analysis of positive and negative emotion frequency scores for computer-as-a-communication-tool and computer-as-an-interactive-object scenarios.The number of factors retained was determined by parallel analysis.(a & b) Results of the emotional experience factor analysis for computer-as-a-communication-tool.(c & d) Results of the emotional experience factor analysis for computer-an-interactive-object.(e & f) Correlation between positive and negative emotional factors, respectively."CT" is short for "computer as a communication tool" and "IO" is short for "computer as an interactive object.".

Table 1 .
The included emotion items.

Table 2 .
Representative HCI and HHI scenarios listed in the questionnaires.