Message Perception within Context-Aware Recommender Systems

Download Message Perception within Context-Aware Recommender Systems

Post on 17-Jan-2016




0 download

Embed Size (px)


Thus far implementation of Context Aware Recommender Systems have primarily focused on what to recommend by deriving results from patterns of behavior and environment to determine optimum product selection for recommendation. Our experiment demonstrates that a purchasers affective state also has an effect on their perception of information presented via a mobile device. We posit that the how and when to recommend are important considerations that have not been fully addressed when considering the display of recommendations. Together with user behaviors associated with purchasing traits, e.g. impulse buying, we explore the information processing styles of mental imagery and analytical processing; risk acceptance; involved user effort; and marketing techniques of positive and negative appeals. Results show that these different methods of presenting information to the purchaser will be successful in obtaining a positive user perception within different affective states. Together an understanding of these information presentation and processing techniques is used to build a representation of a purchasers perception that could be used in m-commerce systems.


<ul><li><p>Message Perception within Context-Aware Recommender Systems </p><p>Mark A. Hooper, Paul Sant </p><p>University of Bedfordshire, </p><p>Department of Computer Science and Technology, </p><p>University Square, Luton, UK, LU1 3JU </p><p>, </p><p>ABSTRACT </p><p>Thus far implementation of Context Aware </p><p>Recommender Systems have primarily focused </p><p>on what to recommend by deriving results </p><p>from patterns of behavior and environment to </p><p>determine optimum product selection for </p><p>recommendation. Our experiment </p><p>demonstrates that a purchasers affective state </p><p>also has an effect on their perception of </p><p>information presented via a mobile device. We </p><p>posit that the how and when to recommend </p><p>are important considerations that have not been </p><p>fully addressed when considering the display </p><p>of recommendations. Together with user </p><p>behaviors associated with purchasing traits, </p><p>e.g. impulse buying, we explore the </p><p>information processing styles of mental </p><p>imagery and analytical processing; risk </p><p>acceptance; involved user effort; and </p><p>marketing techniques of positive and negative </p><p>appeals. Results show that these different </p><p>methods of presenting information to the </p><p>purchaser will be successful in obtaining a </p><p>positive user perception within different </p><p>affective states. Together an understanding of </p><p>these information presentation and processing </p><p>techniques is used to build a representation of </p><p>a purchasers perception that could be used in </p><p>m-commerce systems. </p><p>KEYWORDS </p><p>Recommender systems, personalization, user </p><p>interfaces, affective computing, context-aware </p><p>1 INTRODUCTION </p><p>Research is beginning establish an under-</p><p>standing of user affective, social and phys-</p><p>ical states and their relevance within con-</p><p>text-aware systems [1]. However it is only </p><p>now with the advance of smart-phone sen-</p><p>sor technology that research can truly lev-</p><p>erage this knowledge within the area of </p><p>mobile recommender systems [2]. Though </p><p>research into context-aware recommender </p><p>systems is now showing positive results </p><p>through multi-criteria evaluation of both </p><p>user generated content and environmental </p><p>context the utilisation of contextual infor-</p><p>mation is still thus far limited. </p><p>The focus of this paper is to demonstrate </p><p>that user context can be used to understand </p><p>how an individual reacts to information </p><p>presentation styles via a mobile device. We </p><p>posit that understanding user behavior </p><p>within context is critical to fully realise the </p><p>potential for recommender system results </p><p>through message customisation, especially </p><p>within the developing area of m-commerce </p><p>environments. To support this we define </p><p>and partly verify a framework for recom-</p><p>mender system personalisation that intro-</p><p>duces a new layer of system intelligence </p><p>through the use of message customisation </p><p>based on user contextual behavior. </p><p>We discuss the theory that mood and emo-</p><p>tions influence our selection of cognitive </p><p>processing modes which in turn provide an </p><p>Proceedings of the Third International Conference on E-Technologies and Business on the Web, Paris, France 2015</p><p>ISBN: 978-1-941968-08-6 2015 SDIWC 59</p></li><li><p>insight into the level of message persua-</p><p>sion. To develop our perception trait model </p><p>we have developed hypotheses that focus </p><p>on the relationships between affective </p><p>states, cognitive capacity and behavior. It </p><p>is generally agreed that positive moods re-</p><p>sult in reduced capacity and therefore a </p><p>favouring towards heuristic processing, </p><p>whereas negative moods can facilitate </p><p>more complex detail analysis [3]. </p><p>Different affective states can also influence </p><p>different purchaser traits including, moti-</p><p>vation [4], [5], impulse buying [6], com-</p><p>pulsive buying [7], brand attitude and ad-</p><p>claim recall [8], risk-taking and self-image </p><p>[9]. Myers and Sar [8] provide valuable </p><p>insight into how a pre-existing mood af-</p><p>fects a users response to imagery inducing </p><p>advertisements. We show that understand-</p><p>ing these cognitive ability and behaviors </p><p>should strengthen recommendation con-</p><p>version when coupled with standard rec-</p><p>ommender techniques. </p><p>The rest of this paper is structured as fol-</p><p>lows. We investigate a number of affect </p><p>behavior relationships and their affect user </p><p>perception in section 2. We then discuss </p><p>our implementation of an Android applica-</p><p>tion used to capture in the wild user per-</p><p>ception of specific messaging styles, see </p><p>section 3. In section 4 we present and ana-</p><p>lyse our results and in section 5 we discuss </p><p>limitations and opportunities for further </p><p>research. Section 6 presents our final con-</p><p>clusions. </p><p>2 AFFECTIVE PURCHASING BEHAVIOR </p><p>2.1 Consumer Behavior and Advertisement Techniques </p><p>We hypothesise that understanding behav-</p><p>ior towards a set of situational contexts can </p><p>be utilised to optimise context-aware sys-</p><p>tems by providing a reasoned reaction to-</p><p>wards, not only the presented options, but </p><p>also the method of presentation to the user. </p><p>We stipulate that the addition of affective </p><p>phenomena to the contextual picture is to </p><p>also consider the users behavior as reac-</p><p>tional and not just as an additional element </p><p>of the context that influences preferences. </p><p>We can thus potentially indicate behavior </p><p>towards the advertisement content and the </p><p>medium (i.e. text, image or video), as dis-</p><p>cussed in the paper by [10]. This hypothe-</p><p>sis leads us to consider behavior as a key </p><p>concept to advance research within Con-</p><p>text-Aware Recommender Systems (CARS), </p><p>thus providing further potential for solu-</p><p>tions to commercial recommender system </p><p>that operate in complex environments tar-</p><p>geting audiences with distinct catalogue </p><p>product types numbering in their millions. </p><p>Though an everyday occurrence the act of </p><p>purchasing an item, whether in store or on-</p><p>line, is a complex process that includes </p><p>both environmental factors and consumer </p><p>characteristics, marketing and environment </p><p>stimuli, motivation and personality factors. </p><p>There are many drivers that form an indi-</p><p>viduals approach to the purchasing cycle. </p><p>These complex emotional drivers include </p><p>social potency and closeness, stress reac-</p><p>tion, control, harm avoidance, traditional-</p><p>ism, and absorption [6], enjoyment [11], </p><p>and perception of risk [12]. These in turn </p><p>influence purchasing behaviors of impulse </p><p>[6], need for convenience and information </p><p>search [13]. Personality traits generally </p><p>form our emotional responses to situations </p><p>so are key to understanding particular pur-</p><p>chasing behaviors such as impulse buying </p><p>[6]. As its name implies, impulse buying is </p><p>an unplanned event that is made through a </p><p>snap judgment process. By reviewing </p><p>stimuli to form a quick, convenient repre-</p><p>sentation of a situation it is often character-</p><p>ized as a type of holistic processing that </p><p>has advantages of speed, and reduced cog-</p><p>Proceedings of the Third International Conference on E-Technologies and Business on the Web, Paris, France 2015</p><p>ISBN: 978-1-941968-08-6 2015 SDIWC 60</p></li><li><p>nitive effort [14]. </p><p>A typical example of a holistic processing </p><p>technique is mental imagery, this is an in-</p><p>fluential tool for advertisers for enhancing </p><p>brand attitudes while engaging consumers </p><p>[8]. The process not only includes the mar-</p><p>keting message cues of visual, auditory, </p><p>tactile and emotional [15], but also draws </p><p>upon the purchasers previous experience, </p><p>memories and daydreams to fully form a </p><p>visual image of the situation [13]. This </p><p>contrasts with analytical processing which </p><p>forms a comprehensive understanding of a </p><p>situation through analysis of individual </p><p>stimulus characteristics. Burroughs [14], </p><p>determines that the style of processing is </p><p>selected depending on the characteristics </p><p>of task, stimulus and the individual con-</p><p>sumer. </p><p>An individuals purchase behavior can be </p><p>predicted through their perception of risk, </p><p>a consumer will avoid impulse buying </p><p>when perception of risk is high [16]. </p><p>Bhatnagar et al. [12] report on relation-</p><p>ships between risk, convenience and on-</p><p>line shopping stating that certain product </p><p>categories. Music and CDs, are not gener-</p><p>ally considered risky because of the practi-</p><p>calities of shopping on-line, i.e. reduction </p><p>of costs and an increase in convenience to </p><p>make purchases more likely [12]. Products </p><p>with higher value are perceived as to have </p><p>a higher risk, however they could be </p><p>viewed as being more convenient to be </p><p>purchased on-line if more involved [12], or </p><p>are likely to require an evaluation process </p><p>or other pre-purchase activity [13]. </p><p>Evaluation processes used in information </p><p>search rely upon analytical information </p><p>processing to produce a comprehensive </p><p>understanding [14]. Information search via </p><p>the use of mobile phones is important in </p><p>the evaluation of alternatives and pre-</p><p>purchasing activities, e.g. finding discount </p><p>vouchers [13]. Using an analytical pro-</p><p>cessing style the individual will attempt to </p><p>understand details of the purchasing situa-</p><p>tion from all angles, in doing so they will </p><p>be more likely to identify all important in-</p><p>formation including negative factors and </p><p>therefore be able to limit risky conse-</p><p>quences [14]. </p><p>In addition to considering styles of infor-</p><p>mation processing, risk acceptance and </p><p>levels of processing effort we should also </p><p>understand how common techniques for </p><p>manipulating emotions are important in </p><p>marketing campaigns. We have briefly </p><p>mentioned emotional drivers that shape our </p><p>decisions and behavior, the use of emo-</p><p>tional appeals in marketing create a psy-</p><p>chological reaction that could be resolved </p><p>by acting upon the appeal message, e.g. </p><p>through purchasing an item [17]. Fear ap-</p><p>peal has been widely used in commerce </p><p>and awareness campaigns with varied suc-</p><p>cess depending on content and severity of </p><p>message [17], however the basic premise is </p><p>to focus on insecurity and concerns in or-</p><p>der to prompt action. Positive appeals also </p><p>exist and are written to engage arouse </p><p>emotions like love, desire or humour to </p><p>invoke behaviors including self-esteem </p><p>[18]. So we can summarize the above by </p><p>identifying four categories that help form </p><p>knowledge of consumer engagement with </p><p>marketing messages and around which we </p><p>can build our hypotheses: </p><p> Processing style mental imagery vs. analytical </p><p> Risk acceptance low risk vs. high risk </p><p> Cognitive capacity low effort vs. high effort </p><p> Appeal type positive vs. negative appeals </p><p>Proceedings of the Third International Conference on E-Technologies and Business on the Web, Paris, France 2015</p><p>ISBN: 978-1-941968-08-6 2015 SDIWC 61</p></li><li><p>2.2 The Influence of Emotion upon Consumer Behavior </p><p>This section discusses a number of hypoth-</p><p>eses that together will represent a broad </p><p>understanding of user perception (and thus </p><p>potential user behavior) for use with </p><p>CARS. We expect that by determining a </p><p>users affective state as being positive we </p><p>will be able to establish a different set of </p><p>likely behaviors when compared to a nega-</p><p>tive state. This notion would then support a </p><p>systems approach in presenting certain </p><p>information or taking a specific action. </p><p>Myers and Sar [8] discuss the relevance of </p><p>mood and its likelihood as a context for an </p><p>advertisement to be successful. Alongside </p><p>previous research efforts they state that </p><p>their findings appear to show that positive </p><p>evaluation of an advert is enhanced when </p><p>in a positive mood through the increased </p><p>ability to undertake mental imagery pro-</p><p>cessing. They also suggest that capacity to </p><p>evaluate detailed information is reduced </p><p>during periods of positive mood but this </p><p>then increases during periods of negative </p><p>mood. This is supported by Escalas [19], </p><p>who notes that the effort in generating the </p><p>mental imagery decreases the ability to un-</p><p>dertake further cognitive tasks such as crit-</p><p>ically analyse the adverts content which </p><p>could in turn produce more negative evalu-</p><p>ations. </p><p>These findings suggest that mood is a use-</p><p>ful context when ascertaining how to pre-</p><p>sent items via a recommender system. The </p><p>use of mental imagery may act to make the </p><p>recommendation more appealing as mood </p><p>positivity increases and thus conducive to </p><p>the actual success of the advert. Where a </p><p>negative mood is present and mental im-</p><p>agery deemed less favourable then recom-</p><p>mender messages that provide detail suited </p><p>to analytical processing could be more </p><p>successful. Presenting recommender items </p><p>through the use of mental imagery or ana-</p><p>lytical processing depending on the users </p><p>affective context are a novel concepts, </p><p>therefore we posit that: </p><p>H1: Processing Style </p><p>H1a: that a positive correlation will be </p><p>achieved between user affective state and </p><p>the perception of a mental imagery induc-</p><p>ing statement </p><p>H1b: that a negative correlation will be </p><p>achieved between user affective state and </p><p>the perception of a statement using analyt-</p><p>ical, detail-oriented reasoning </p><p>The relationship between risk-taking and </p><p>mood holds a similar theme. Research has </p><p>often reported that when we are in a posi-</p><p>tive mood and are presented with a hypo-</p><p>thetical situation we are more risk favoura-</p><p>ble. For example Yuen and Lee [20], note </p><p>that those in a positive mood are less con-</p><p>servative and more open to risk. However </p><p>they do report significant differences of the </p><p>effect of mood on levels of risk ac-</p><p>ceptance. This could be explained by not-</p><p>ing Isen [21], who suggests that when a </p><p>person is presented with a real risk situa-</p><p>tion they are more likely to be risk adverse. </p><p>Therefore, along with other research such </p><p>as [22] we postulate that negative mood is </p><p>more complex than basic categories of la-</p><p>boratory induced moods of sad as used </p><p>by [20]. In addition to this it may also be </p><p>logical to suggest that real life situational </p><p>mood and emotions may potentially pro-</p><p>duce different results to laboratory find-</p><p>ings, especially under different situational </p><p>contexts. </p><p>Previous research has also determined the </p><p>effect of mood on our perception of risk. </p><p>Lee [16], presents results that demonstrate </p><p>that elements of positive mood are related </p><p>to impulsive buying traits. Brave and Nass </p><p>[9] state that it is expected that we will en-</p><p>deavour to maintain a positive sensation by </p><p>Proceedings of the Third International Conference on E-Technologies and Business on the Web, Paris, France 2015</p><p>ISBN: 978-1-941968-08-6 2015 SDIWC 62</p></li><li><p>being more risk adverse with engagement </p><p>likely to continue with low-risk impulse </p><p>sales. In addition to this, when in a nega-</p><p>tive state we are generally aiming to recap-</p><p>ture a more positive outlook and are more </p><p>likely to engage with riskier purchases to </p><p>kick-start the positive emotional process </p><p>[9]. We follow this reasoning for the next </p><p>hypothesis pair. </p><p>H2: Risk Acceptance </p><p>H2a: that a positive correlati...</p></li></ul>


View more >