The Prefiltering Techniques in Emotion Based Place …downloads.hindawi.com/journals/acisc/2017/5680398.pdf · 2019-07-30 · The Prefiltering Techniques in Emotion Based Place Recommendation

Research ArticleThe Prefiltering Techniques in Emotion Based PlaceRecommendation Derived by User Reviews

U A Piumi Ishanka and Takashi Yukawa

Graduate School of Engineering Nagaoka University of Technology 1603-1 Kamitomioka-machi Nagaoka Japan

Correspondence should be addressed to U A Piumi Ishanka s135063stnnagaokautacjp

Received 2 August 2017 Accepted 19 September 2017 Published 22 October 2017

Academic Editor Shyi-Ming Chen

Copyright copy 2017 U A Piumi Ishanka and Takashi YukawaThis is an open access article distributed under the Creative CommonsAttribution License which permits unrestricted use distribution and reproduction in any medium provided the original work isproperly cited

Context-aware recommendation systems attempt to address the challenge of identifying products or items that have the greatestchance of meeting user requirements by adapting to current contextual information Many such systems have been developed indomains such as movies books and music and emotion is a contextual parameter that has already been used in those fields Thispaper focuses on the use of emotion as a contextual parameter in a tourist destination recommendation system We developed anew corpus that incorporates the emotion parameter by employing semantic analysis techniques for destination recommendationWe review the effectiveness of incorporating emotion in a recommendation process using prefiltering techniques and show that theuse of emotion as a contextual parameter for location recommendation in conjunction with collaborative filtering increases usersatisfaction

1 Introduction

Context-awareness has been introduced into recommenda-tion systems to ensure that both long-term and short-termuser needs are recognized by considering not only preferencehistory but also the current situation This is because if therecommendation process only relies on preference historyit may not correctly discard an isolated situation such as agift purchase (ie if a user buys a gift for a friendrsquos childa baby suit user receives suggestions for the baby itemsrepeatedly) because it cannot adapt to the current situationContext-aware recommendation systems (CARS) thereforeincorporate contextual information including location timeand activity and even advanced parameters such as emotionand personality

Previous studies of recommendation systems have inves-tigated the incorporation of emotion into the recommen-dation process with a particular emphasis on the impactof emotion and personality traits on the human decision-making process The role of emotion has been identified indifferent stages of the content consumption process including(i) the entry stage (ii) the consumption stage and (iii) theexit stage while the consumption stage is highly influenced

by emotion [1] Emotions are mental states usually caused byan event of importance to the subject and have beenmodeledin various studies The universal model classifies emotioninto fixed categories such as Ekmanrsquos six basic emotions(happiness anger fear sadness disgust and surprise) [2] andPlutchikrsquos eight basic and prototypical emotions (joy sadnessanger fear trust disgust surprise and anticipation) [3] Inthe dimensionalmodel each emotion is described as a point ina continuous multidimensional space where each dimensionrepresents a quality of an emotion The most frequently useddimensions are valence arousal and dominance [1]

According to various psychological studies a place canhave an impact on memories sentiments and emotionalwell-being [4] Debored [5] introduced the concept ofpsychogeography and tracked the influence of geographicalenvironments on the emotions and behavior of individualsAlthough location is commonly analyzed by consideringcontext and its impact on emotional well-being no taxonomyhas been derived for the emotion that a location may evoke[6] Moreover it is vital that any dataset contains affectiveparameters Studies of music and movie recommendationsystems datasets have used the LDOS-CoMoDa LJ2MmyPersonality LDOS-PerAff-1 DEAP ANET IADS ANEW

HindawiApplied Computational Intelligence and So ComputingVolume 2017 Article ID 5680398 10 pageshttpsdoiorg10115520175680398

2 Applied Computational Intelligence and Soft Computing

Predictions

Recommendations active users

Items

UsersRatings table

+

TopmdashN list of items for

j for active userPajmdashprediction on item

I1 I2 I3 Ii ImU1

U2

Ua

Un

middot middot middotmiddot middot middot

Figure 1 Collaborative filtering process

IAPS and 1000 songs datasets [7] but none of these relate totourist destination recommendation data

In development of the proposed system we addressed thepaucity of datasets with contextual parameters by employingSentimental Analysis (SA) to acquire the emotional states ofusers based on user reviews for 100 destinations We thenderived emotional tags and manually compared the accuracyof the tags for each of the destinations

In this study a CARS that uses partial contextual userpreferences in the form of user item context and ratingwas developed This is unlike a traditional recommendationsystem which is based on knowledge of preferences of aset of items and the input data is in the form of user itemand ratingThe contextual information that we consider herecan be applied to various stages of the recommendationprocess A CAR process based on contextual user prefer-ences elicitation and estimation can take one of three formsas follows contextual prefiltering contextual postfilteringor contextual modeling In this study we implemented acontextual prefiltering paradigm-based solution and use thecontextual information to select themost relevant item times userdata for generating recommendations [8] The prefilteringapproach allows for the deployment of any of the numeroustraditional recommendation techniques previously proposed[9]

The main objectives of this study are as follows

(1) Deriving suitable datasets for tourist destination rec-ommendations considering emotion as a contextualparameter

(2) Examining how emotion influences recommendationof destinations using prefiltering techniques

2 Background

The body of literature related to recommendation systemsis constantly expanding Previous studies have introducedalgorithms for recommendations improved ways of buildinguser models to represent user preferences interests andbehaviors and domain-specific applications [10] Petrevskaand Koceski (2012) proposed a tourism recommendationsystem based on user preferences interests and desires andsuggested tourist attractions based on profiling user behav-iors such as reading other reviews tomake decisions [11] Des-tinations were observed as objects and changes in user pref-erences following subsequent visits tracked according to therating and user behavior was considered in the profiling and

recommendations generated based on collaborative filteringSarkaleh et al (2012) suggested a tourism recommendationsystem using location and user features as contextual param-eters [12] Zheng et al (2012) proposed a CARS for travel rec-ommendations using differential contextual relaxation andconsidered trip type duration originality destination cityand month as contextual parameters with user collaborativefiltering to propose a hybrid recommendation to mitigate thedisadvantages of the solo use of prefiltering and postfilteringtechniques [13] De Pessemier et al (2015) suggested a systemwith group recommendation for traveler destinations basedon a userrsquos rating profile personal interests and specificationfor their next destination by following a hybrid approachin combination with content-based collaborative filteringand knowledge-based strategies [14] Collaborative filteringis a widely adapted recommendation algorithm in whichpredictions and recommendations are based on the ratingsor the behavior of the users in the system [15 16]

The approach considers a list of n number of users 119880 andlist of 119898 number of items 119868 As shown in Figure 1 the user119906119886 119906119886 isin 119880 is called the active user for whom the task of thecollaborative filtering algorithm is to find an item likelinessthat can be in two forms 119875119886119895 is the predictive rating for item119894 for the active user 119880119886 where 119895 isin 119898 The prediction rating isa numerical value with the same scale as the rating providedby user 119906119886The recommendation is the list of119873 items that theactive user will like the most [17]

Collaborative filtering techniques can be either user-based or item-based The user-based collaborative filteringmodel recommends items based on computing similar neigh-bors and creates a group of users that are compatible with atarget user Item-based collaborative filtering computes sim-ilarity based on items and finds items that are similar to thegiven userrsquos rated items [18] A traditional recommendationsystem starts by estimating the rating function R using aninitialized set of ratings that are either explicitly provided byusers or are implicitly inferred by the system for user 119906 anditem 119894 that has not been rated

R user times item 997888rarr rating (1)In a CARS preferences are predicted by incorporating

available contextual information into the recommendationprocess and R is estimated by

R user times item times context 997888rarr rating (2)where user and item are the space of users and items andrating is the space of rating for the user and item pairs The

Applied Computational Intelligence and Soft Computing 3

Data Filtered data 2D recommender Contextual recommendation

C U

i1 i2 i3U times I rarr RU times I times C times R U times I times R

Figure 2 Prefiltering process

contextual information can be applied during various stagesof the recommendation process and the form of the context-aware recommendation can be contextual prefiltering (con-textualization of input) contextual postfiltering (contextual-ization of output) or contextual modeling (contextualizationof recommendation function) [8] In contextual prefilteringthe contextual information is used to select the relevant setof ratings and ratings are predicted for active users using anytraditional recommendation functionThe context is then setas a query for filtering relevant ratings (see Figure 2)

The most appropriate use of prefiltering technique canvary with the application One approach is to use amodel thattargets a local context model for each situation Another is touse generalized prefiltering which allows for the generaliza-tion of the data filtering query based on a specific context [19]Generalized prefiltering states the ratings based on the relatedcontextual situations and derives a collection of predictionmodels based upon the ratings for each segment

Affective computing and SA are combined in variousresearch fields and many of advanced SA techniques havebeen developed and many commercial and academic toolssuch as IBM (httpswwwibmcomanalytics) SAS (httpswwwsascomsocial) Oracle (httpswwworaclecomsocial)SenticNet (httpwwwbusinesssenticnet) and Luminoso(httpswwwluminosocom) emerged for facilitating polar-ity evaluations and or mood classification though most ofthem are costly and highly limited to set of emotions [20]SA is considered as computational treatment of opinionssentiments and subjectivity of texts while opinion miningis a tedious task since a comprehensive knowledge of most ofthe explicit and implicit regular and irregular and syntacticand semantic rules of a language is to be considered in theprocess [21 22]

In the state of the art of sentiment classification tech-niques the lexicon approaches rely on a sentiment lexiconand the classifications methods using machine learning canbe further divided into supervised and unsupervised learn-ing methods The supervised learning methods comprisedmany classifiers including probabilistic classifier naive Bayesclassifier linear classifier Support Vector Machine classi-fiers Neural Network decision tree classifiers rule-basedclassifiers and meta-based classifiers [21] Currently the SAtechniques are enriched with deep learning approaches likeDeep Convolutional Neural Networks as well The Convolu-tional Neural Networks (CNNs) are very similar to ordinaryNeural Networks (NN) the main difference is the numberof layers where CNN are just several layers of convolutionswith nonlinear activation functions applied to the resultswhile in traditional NN each input neuron is connectedto each output neuron in the next layer In CNNs insteadconvolutions are used over the input layer to compute the

output This results in local connections where each regionof the input is connected to a neuron in the output Eachlayer applies different filters typically hundreds or thousandsand combines their results [23 24] A drawback of CNN as aclassifier is that it finds only a local optimum since it uses thesame backpropagation technique as MLP [25]

3 Incorporating Emotion intoPlace Recommendation

31 Emotion Tag Acquisition The main challenge with aCARS domain is the lack of appropriate contextual datasetsThe LDOS-PerAff-1 corpus is one of the datasets createdto fulfill this issue [26] It incorporates video clips of usersresponding to emotional stimuli and ratings with personalitytraits Data acquisition was performed by presenting a setof images and asking subjects to rate the images as ifthey were choosing images for their computer wallpaperThe LDOS-CoMoDa corpus is another dataset introducedas a context-aware movie dataset comprising 12 contextualdimensions with 2291 ratings rated by 121 users on 1232items Among the contextual parameters suggested threeemotional dimensions were included endEmo emotionalstates at the end of the movie domEmo the emotional stateexperienced most during the movie and mood the emotionof the user when they are watching the movie [27] Theacquisition of emotion has been investigated using a varietyof technologies such as the detection of facial expressionsemotion inference from sensors and other approaches basedon voice speech body language and postures The difficultyin applying these techniques in the recommendation systemdomain is the complexity in adapting these techniques tothe system implementation Therefore previous studies havefocused on emotion states inferred from reviews by using SAtheory SA techniques can be used to extract emotion fromreview texts including joy sadness fear anger and surprise[21] Emotion detection can be implemented using machinelearning or lexicon-based approaches with the latter beingmore frequently used [27]

In this study a context-aware dataset was derived bycollecting data for the global top 100 tourist attractions in2016 The data including description location and imageswere obtained from Wikipedia (httpsenwikipediaorg)while the average rating and 100 user reviews for each destina-tionwere collected fromTripAdvisor (httpswwwtripadvisorcom) The reviews were analyzed and classified to acquireemotion tags to represent user emotional states for a place intwo stages First we usedToneAnalyzer (httpstone-analyzer-demomybluemixnet) which measures emotional tone toget a sense of the overall tone of the review (joy fear sad-ness disgust and anger) Second we expanded the positive


Table 1 Sample results of tone analysis created by Tone Analyzer

Place ID Anger Disgust Fear Joy Sadness10 0073451 0114556 0111547 0530747 023410711 0074452 0115561 0109812 0501336 025950912 0082424 0065996 0063544 0617 026585613 0091267 0115791 0120392 0506598 023192614 0151867 0076331 0070043 0567042 026799215 0135269 0064383 0101094 0545255 026085716 0142556 0161189 0165065 0422934 031923917 0095104 006126 0152121 0513339 026493118 0123192 0102279 0113161 0524549 036032719 0092798 0086266 0113484 0526654 024062420 0067278 0072824 0074158 0611672 0285493

Table 2 Extract from the emotion lexicon used in the classification

Joy Surprise Anticipation Trust Fear Anger Sadness DisgustAbsolution Abandonment Abundance Abacus Abandon Abandoned Abandon AberrationAbundance Abduction Accelerate Abbot Abandoned Abandonment Abandoned AbhorAbundant Abrupt Accolade Absolution Abandonment Abhor Abandonment AbhorrentAccolade Accident Accompaniment Abundance Abduction Abhorrent Abduction AbjectAccompaniment Accidental Achievement Academic Abhor Abolish Abortion AbnormalAccomplish Accidentally Acquiring Accolade Abhorrent Abomination Abortive AbominableAccomplished Accolade Addresses Accompaniment Abominable Abuse Abscess AbominationAchieve Advance Adore Accord Abomination Accursed Absence AbortionAchievement Affront Adrift Account Abortion Accusation Absent AbundanceAcrobat Aghast Advance Accountability Absence Accused Absentee Abuse

emotion scale from ldquojoyrdquo to ldquojoy anticipation trust andsurpriserdquo and employed SA techniques Text preprocessingtechniques were used to derive emotion tags from reviewsfor each location A list of words indicative of each emotionwas used The emotion lexicon from the National ResearchCouncil (NRC) Canada which is based on Plutchikrsquos eightemotions and two negative and positive sentiments [28] wereused to calculate term frequencies

The lexicon set derived by NRC comprises unigrams andbigrams of the Macquarie Thesaurus [29] all terms in theGeneral Inquirer and the WordNet-Affect Lexicon Thusthe used lexicon set comprises unigrams and bigrams of theMacquarie Thesaurus 800 and 787 lexicons as adjectivesadverbs nouns and verbs respectively all the terms inthe General Inquirer 8132 lexicons as negative positiveand neutral and the WordNet-Affect Lexicon 640 lexiconsrepresenting emotion categories anger disgust fear joysurprise and sadness

In the Tone Analyzer analysis for each emotion a scoreof less than 05 indicates that the emotion is unlikely tobe perceived in the content and a score greater than 075indicates a high likelihood that the emotionwill be perceivedThe overall results show that the highest tone value wasrecorded in the joy group for all locations and 88 of thejoy group indicates that the emotion is likely to be perceivedin the selected review texts (see Table 1)

Text preprocessing is a significant task in text miningtechniques and its application is the first step in any systemThe main aim behind it is to represent each document as afeature vector so it separates the text into individual wordsThe quality of the classification process is highly dependenton this feature selection process Therefore it is important toselect meaningful keywords and discard words that do notenable distinguishing of the documents [30]

Next we performed a term frequencies (TFs) calculationbased on the lexicons of the eight emotion groups Table 2illustrates the part of the lexicon used in classification Wechose the emotion category for a destination based upon themaximum frequency value as shown in Table 3The TFs werecalculated for each review and the total TFs were calculatedfor each emotion We then selected the emotion tag for eachdestination based on the highest frequency (see Table 4)

In the case of the occurrence of multiple emotionswe assumed that if a negative emotion category and apositive emotion category appeared with equal frequencythe negative one was stronger This assumption is based onthe fact that humans are more honest when stating negativeemotions In terms of the positive emotions occurring withan equal frequency we assumed that joy trust anticipationand surprise are stated in a descending order by consideringhow humans tend to identify and state these emotions (seeTable 4)


Table 3 Emotion tag derivation based on TF values

Place ID User ID An Ant D F J S Su T1 1 1 3 1 0 4 0 2 31 2 0 1 0 0 0 0 0 11 3 1 3 1 1 3 2 2 21 4 1 0 1 0 1 0 0 11 5 1 3 0 2 1 1 1 1

Emotion word sum 4 10 3 3 9 3 5 81 Emotion tag AnticipationAn angry Ant anticipation D disgust F fear J joy S sadness Su surprise and T trust

Table 4 Emotion tag derivation of TFs

Place ID J Su T Ant D F An S Emotion tag1 29 14 55 42 6 12 13 16 Trust2 28 12 32 23 1 10 5 4 Trust3 31 14 21 24 0 3 4 10 Joy4 17 7 17 13 0 6 4 6 Joy5 39 18 44 44 5 14 10 11 Trust6 31 32 43 41 8 11 16 18 Trust7 51 26 41 38 2 3 2 8 Joy8 46 17 48 39 9 11 10 9 Trust9 24 13 22 22 2 8 6 11 Joy10 23 11 28 21 1 6 4 5 Trust11 45 19 49 36 5 12 4 7 Trust12 44 31 26 64 4 7 7 5 Anticipation13 18 7 20 16 1 5 4 4 Trust14 49 21 47 62 6 14 7 9 Anticipation15 35 16 37 43 2 21 7 12 AnticipationAn angry Ant anticipation D disgust F fear J joy S sadness Su surprise and T trust

PreprocessingReviewcollection

Stop wordremoval TF calculation

Stop word list NRC lexicons

Emotion tagcreation

Figure 3 Emotion tag creation

We collected 9998 ratings from 8470 users for the 100selected locations and derived emotion tags for each locationfor each user From this information we derived two datasetsplace dataset 119901119897119886119888119890 119894119889 119899119886119898119890 119889119890119904119888119903119894119901119905119894119900119899 119901119897119886119888119890 119888119886119905119890119892119900119903119910119897119900119888119886119905119894119900119899 119886V119890119903119886119892119890 119903119886119905119894119899119892 119890119898119900119905119894119900119899 119905119886119892 anduser dataset 119906119904119890119903119894119889 119901119897119886119888119890 119894119889 119906119904119890119903 119890119898119900119905119894119900119899 119905119886119892 119906119904119890119903 119903119886119905119894119899119892 Figure 3 illus-trates the emotion tag derivation process

119860 is the total number of cases (destinations) to which thesystem assigned emotion tags

119862 is the number of correct cases based on manualjudgment

Precision =119862

119860(3)

We calculated the precision of emotion tag detection andcompared the results manually by reading review texts forthe place dataset The results show that the precision of thedetection process is 633 [31]

32 Recommendation SystemDevelopment We implementedthe proposed recommendation system using the two deriveddatasets and loaded the location data into the database ofthe system The dataset was input to the recommendationfunction based on the emotion state of the user when theylogged in as shown in Figure 4 Using the prefilteringtechniques in CARS the similarity and predictive ratingvalues were calculated for each user and the top five place


Figure 4 Emotion gathering

User

Filtering datasetwith emotion matrix

Predictive ratingcalculation

list

(i) Rate places(ii) Track emotion

Database

Emotion tags

Top N place recommendation

User times place

Figure 5 Recommendation process

recommendations were generated The systemrsquos recommen-dation engine consists of two phases based on collaborativefiltering without emotion (CFN) and collaborative filteringwith emotion (CFE)

In the implementation we used item-item collaborativefiltering to develop and review our contextual parameterson the derived dataset The recommendation process isillustrated in Figure 5

Each recommendation based on emotion was analyzedby considering the three emotion groups derived accord-ing to Plutchikrsquos emotion classification system In the rec-ommendation process data was selected for the locationrecommendation engine based on these three groups fromthe assumption that the recommendation should fall on thepositive emotion scale The results of the Tone Analyzeranalysis reinforced the use of the positive emotion scalebecause the collected reviews were more likely to have apositive emotion Therefore disgust fear anger and sadnesswere rearranged into three positive groups (anticipation joyand trust) based on Plutchikrsquos comprehensive list of eightprimary emotions arranged as opposing pairs (see Table 5)To avoid a negative emotion category fear and anger wereallocated to the joy and trust groups respectively Furtherwe used these three groups to evaluate the influence of useremotion on the recommendation

In the recommendation process a prediction of a targetuserrsquos rating on an unrated target item was calculated byconsidering the userrsquos rating of observed items This allowsfor user-item rating pairs to be used to rate value predictionsas shown in Figure 6 [32]

For item-item collaborative filtering users who haverated both item 119894 and item 119895 are identified and then thesimilarities are computed [33] The similarity calculation isperformed based on measures such as the Pearson corre-lation Euclidean distance Tanimoto coefficient or the log-likelihood similarity In the proposed process the similarity

Table 5 Emotion groups in the recommendation process

Group I Anticipation Anticipation surpriseGroup II Joy Joy sadness fearGroup III Trust Trust disgust anger

Item 1 Item 2 Item 3 Item 4 middot middot middot Item nUser 1 1 3 middot middot middotUser 2 2 5 middot middot middotUser 3 5 3 middot middot middotUser 4 3 1 middot middot middotmiddot middot middot middot middot middot middot middot middot middot middot middot middot middot middot middot middot middot middot middot middot

User m 4 5 5 middot middot middot

Similarity distance

Figure 6 User-item matrix

calculationwas based on the log-likelihood ratio which relieson the statistical similarity between two items or users andyielded a sufficient number of items for the recommendationThe log-likelihood ratio utilizes occurrences related to usersor items such as users or items that overlap and the events forwhich both users and items do not have preferences [34 35]

Prediction algorithms estimate the rating that a userwould provide for a target item [36] For item-based predic-tion the simple weighted average can be used to predict theratings [34] Here we calculated the predictive rating 119875119906119894 byuser 119906 for item 119894 as follows

119875119906119894 =sum119899isin119873 (sim (119906 119899) + 1) times 119877119906119899

sum119899isin119873 (sim (119906 119899) + 1) (4)

where sim(119906 119899) is the similarity between the 119899th item anduser 119906 and 119877119906119899 is the rating by user 119906 of item 119899 for all 119873number of items that are based on the Mahout item-basedrecommendation algorithm [37] The similarity calculations


1 0 1 11 1 0 11 0 0 11 0 0 1

Sample user behavior actionsS1 S2 S3 S4

U1

U2

U3

U4


1 075 066 08075 1 05 06

1 04 1 051 05 1 075

Log-likelihood similarity valuesS1 S2 S3 S4

U1

U2

U3

U4

Figure 8 Similarity values

ranged from minus10 to 10 and to avoid negative values weadded 10 to similarity values so that the similarity rangesfrom 00 to 20 The top-five-places recommendation list wascreated based on the highest similarity values from the mostsimilar places set from the places pool Figure 7 illustratesthe sample user behavior action for four users while Figure 8shows the sample log-likelihood values for four users and fourplaces Thus predictive rating calculation is illustrated belowby using an example of user-itemmatrix and similarity valuesin the recommendation process of user 1 for place 1 (11987511990611199041)and user 1 for place 2 (11987511990611199042)

33 Predictive Rating Calculation (Item-Item Collaboration)

11987511990611199041

=((1 + 1) lowast 1 + (1 + 075) lowast 0 + (1 + 066) lowast 1 + (1 + 08) lowast 1)

(1 + 075 + 066 + 08)

=(2 + 0 + 166 + 18)

321= 17

11987511990611199042


(075 + 1 + 05 + 06)

=(175 + 2 + 0 + 16)

285= 188

(5)

Figure 9 shows an example of top-five-places list providedfor a user

34 Evaluation The Travel Destination location recommen-dation system was presented to 16 users Each user wasasked to evaluate two recommended lists according to theuserrsquos preference for each location and the overall preferencefor the list according to the userrsquos current emotion andoverall satisfaction based on the five-point Likert scale The

Table 6 Precision and recall matrix

Selected Not selected TotalRelevant Nrs Nrn Nr

Not relevant Nis Nin Ni

Total Ns Nn N

Figure 9 Recommendation list

evaluation was performed to assess the usersrsquo opinion ofthe quality of the CFE recommendation algorithm so thata baseline comparison with the CFN algorithm could beperformed

Weused the precision andmean average precision (MAP)values of the two approaches in our evaluation Also a 119905-test analysis was performed to test the superiority of CFEagainst the baseline approach The 119905-test evaluates 119905mean ofboth average precision (AveP) values and average preferenceratings (APR) based on preferred and preferred much userratings in the five-point Likert scale Moreover we evaluatedthe recommendation list by considering the emotion groupsderived at the recommendation engine design stage to trackhow the lists correspond to user emotions The overall usersatisfaction of the recommendation systemwas also analyzedUsers were asked to input their emotion from the emoticonscale and evaluate the two lists of five locations

4 Results and Discussion

We used the classification accuracy measure precision inour evaluation This requires a binary do not recommendselect recommendselect scale so we assumed that ratingsof 4 and 5 were good recommendations [38]

Based on the precision recall matrix (see Table 6) preci-sion is stated as follows

precision =119873119903119904119873119904

(6)

The precision values for the CFN and CFE for the 16 userswere calculated as below and the mean precision values forthe CFE were greater compared with CFN Average precisioncalculates the precision at the position of every correct item inthe ranked results list of the recommenderThemean of theseaverage precisions across all relevant lists is the mean averageprecision (MAP)TheMAP is also greater for CFE comparedto CFN (see Table 7)

Ave119875 =sum119899119896=1 (119875 (119896) times rel (119896))

number of relevant items

MAP =sum119902119902=1 Ave119875 (119902)

119876

(7)


Table 7 Precision and mean average precision values

Algorithm Precision Mean av precision CFN 5969 644CFE 6531 738

644946

552100

833

0 20 40 60 80 100 120CFN

CFE_AnticipationCFE_Joy

CFE_TrustCFE_Overall

Mean average precision ()

Alg

orith

m

Figure 10 Mean average precision values with emotion groups

In the above 119875(119896) is the precision at 119896th element rel(119896) is1 if the 119894th item of the list is relevant and119876 is the total numberof lists

Moreover we analyzed the mean average precision basedon emotional groups (MAPE) for each approach

MAPE =sum119862119888=1sum

119902119902=1 Ave119875 (119902)

sum119862119888=1 119876 (8)

where 119862 is the number of emotion groups based upon threegroups

In Figure 10 CFE Trust CFE Joy CFE Anitcipation andCFE Overall denote the collaborative filtering approach forthe trust joy and anticipation emotional groups respectivelyWe compared the performances of the CF approaches bothwith and without emotions in terms of average precisionvalues and average preference ratingsThe hypotheses are thefollowing

H0 120583119888 = 120583CFN and alternatively H119886 120583119888 = 120583CFN andH119886 120583119888 gt 120583CFN where 120583119888 and 120583CFN are the mean aver-age precision rating of the context-aware and noncon-text collaborative filtering approaches respectively

H0 120583cp = 120583CFN and alternatively H119886 120583cp = 120583CFN andH119886 120583cp gt 120583CFN where 120583cp and 120583CFN are the meanpreference ratings of the context-aware and noncon-text collaborative filtering approaches respectively

Since 119879 (test statistic) lt 119905120572] (critical value) we rejectthe null hypothesis in both cases and conclude that the twopopulation means are different at the 005 significance levelwhile for the alternative hypothesis 120583119888 gt 120583CFN and 120583cp gt120583CFN Therefore the test results show that the differencewith the baseline recommender (CFN) in terms of averageprecision (119901 value = 0509) and average rating preferences(119901 value = 0344) is statistically significant Finally usersrsquofeedback on overall satisfaction with the recommended listas well as their opinion based on their current emotion isshown in Table 8 According to the results 60 of users were

satisfied overall with the recommended lists Further 533acknowledged that the provided list matched their currentemotion

5 Conclusion

In this study we established how emotion can impact thetravel destination recommendation process The use of emo-tion as a contextual parameter for location recommendationin conjunction with collaborative filtering increased usersatisfaction In addition we derived emotion tags for eachlocation based on user reviews to examine how the destina-tion can be effected by emotion in a travel destination recom-mendation systemWhile previous studies have incorporatedemotion into recommendations for predefined indoor placesour study incorporated it onto a recommendation systemfor famous tourist attractions The accuracy of detecting thecorrect emotion tag using the lexicon-based approach was63 However we believe that this can be improved usingother SA approaches

Plutchikrsquos emotion categorization was used to deriveboth emotion tags and the acquisition of the emotions ofusers and the recommendation list incorporated positiveemotion categories Moreover the sensitivity of the emotioncontextual parameter in the recommendation was analyzedbased on the accuracy of the lists for the user

6 Discussion and Future Work

In our approach we focused on deriving an emotion tag foreach destination based on user reviews Basically we derivedthe tags based on Plutchikrsquos emotion categories The lexicon-basedmethods for SA are robust result in good cross-domainperformance and can be easily enhanced with multiplesources of knowledge [39] compared to other approaches

In deriving emotion tags opinion mining and othersemantic analysis techniques can also be used to enhanceaccuracy and one such lexical resource SentiWordNet whichis one dictionary of opinionated terms is used in suchtechniques Also the deep learning approaches and opiningmining techniques explained in Background can also beused to enhance the accuracy of emotion word classificationSenticNet is built based on SentiWordNet lexicon and adaptsHourglass of Emotions In this model sentiments are reorga-nized in four independent dimensions that represent differentlevels of activation In fact in this model affective states arenot classified into traditional emotional categories ratherthey are classified into four concomitant but independentdimensions pleasantness attention sensitivity and aptitude[40]

Although we used exact prefiltering which for the useof traditional recommendation algorithms does not considerany rating acquired in situations even slightly different fromthe targeted one it is proposed that the present system becompared with the context modeling approaches in CARS toallow for an evaluation of the performance of the recommen-dation engine and that the system be extended to incorporate


Table 8 User satisfaction for top-five-places list

Algorithm Overall preference for top 5 Preference with the emotion of user for top 5 places ()CFE 60 5333CFN 4667 mdash

user behavior in the system so as to quantitate the sensitivityof each parameter in the recommendation process

Conflicts of Interest

The authors declare that there are no conflicts of interestregarding the publication of this paper

References

[1] M Tkalcic A Kosir and J Tasic ldquoAffective recommendersystems the role of emotions in recommender systemsrdquo inProc The RecSys 2011 Workshop on Human Decision Making inRecommender Systems pp 9ndash13 October 2011

[2] P Ekman ldquoBasic Emotionsrdquo in Handbook of Cognition andEmotion pp 45ndash60 1999

[3] R Plutchik ldquoA general psychoevolutionary theory of emotionrdquoTheories of Emotion vol 1 no 4 pp 3ndash31 1980

[4] L Castello Rethinking the meaning of place conceiving placein architecture-urbanism Routledge (2016)

[5] G Debord Introduction to a critique of urban geographyCritical Geographies A Collection of Readings (1955)

[6] M Kaminskas and F Ricci ldquoEmotion-based matching of musicto placesrdquo in Emotions and Personality in Personalized Servicespp 287ndash310 Springer 2016

[7] A Odic A Koir and M Tkalcic ldquoAffective and PersonalityCorporardquo in Emotions and Personality in Personalized Servicespp 163ndash178 Springer 2016

[8] GAdomavicius andA Tuzhilin ldquoContext-aware recommendersystemsrdquo in Recommender Systems Handbook pp 191ndash226Springer US 2nd edition 2015

[9] G Adomavicius and A Tuzhilin ldquoToward the next generationof recommender systems a survey of the state-of-the-art andpossible extensionsrdquo IEEE Transactions on Knowledge and DataEngineering vol 17 no 6 pp 734ndash749 2005

[10] G Gonzalez L J De La Rosa M Montaner and S DelfinldquoEmbedding emotional context in recommender systemsrdquo inData EngineeringWorkshop IEEE 23rd International Conferencpp 845ndash852 IEEE 2007

[11] B Petrevska and S Koceski ldquoTourism recommendation systemempirical investigationrdquo Revista de turism-studii si cercetari inturism vol 14 pp 11ndash18 2012

[12] M K Sarkaleh M Mahdavi and M Baniardalan ldquoDesigninga tourism recommender system based on location mobiledevice and user features in museumrdquo International Journal ofManaging Information Technology vol 4 no 2 p 13 2012

[13] Y Zheng R Burke and B Mobasher ldquoDifferential contextrelaxation for context-aware travel recommendationrdquo in Inter-national Conference on Electronic Commerce andWeb Technolo-gies pp 88ndash99 Berlin Germany 2012

[14] T De Pessemier J Dhondt K Vanhecke and L MartensldquoTravelWithFriends a hybrid group recommender system fortravel destinationsrdquo in Workshop on Tourism Recommender

Systems (TouRS15) in Conjunctionwith the 9th ACMConferenceon Recommender Systems (RecSys 2015) pp 51ndash60 2015

[15] P Resnick and H R Varian ldquoRecommender systemsrdquo Commu-nications of the ACM vol 40 no 3 pp 56ndash58 1997

[16] K Goldberg T Roeder D Gupta and C Perkins ldquoEigentastea constant time collaborative filtering algorithmrdquo InformationRetrieval vol 4 no 2 pp 133ndash151 2001

[17] B Sarwar G Karypis J Konstan and J Riedl ldquoItem-based col-laborative filtering recommendation algorithmsrdquo inProceedingsof the 10th international conference onWorldWideWeb pp 285ndash295 ACM 2001

[18] M Tkalcic A Kosir and J Tasic ldquoThe LDOS-PerAff-1 corpusof facial-expression video clips with affective personality anduser-interaction metadatardquo Journal on Multimodal User Inter-faces vol 7 no 1-2 pp 143ndash155 2013

[19] G Adomavicius R Sankaranarayanan S Sen and A TuzhilinldquoIncorporating contextual information in recommender sys-tems using a multidimensional approachrdquo ACM Transactionson Information and System Security vol 23 no 1 pp 103ndash1452005

[20] E Cambria ldquoAffective Computing and Sentiment AnalysisrdquoIEEE Intelligent Systems vol 31 no 2 pp 102ndash107 2016

[21] W Medhat A Hassan and H Korashy ldquoSentiment analysisalgorithms and applications a surveyrdquo Ain Shams EngineeringJournal vol 5 no 4 pp 1093ndash1113 2014

[22] E Cambria S Poria R Bajpai and BW Schuller ldquoSenticNet 4a semantic resource for sentiment analysis based on conceptualprimitivesrdquo in COLING pp 2666ndash2677 2016

[23] M M Lopez and J Kalita ldquoDeep Learning applied to NLPrdquohttpsarxivorgabs170303091

[24] O Araque I Corcuera-Platas J F Sanchez-Rada and C AIglesias ldquoEnhancing deep learning sentiment analysis withensemble techniques in social applicationsrdquoExpert SystemswithApplications vol 77 pp 236ndash246 2017

[25] S Poria E Cambria and A Gelbukh ldquoDeep convolutionalneural network textual features andmultiple kernel learning forutterance-level multimodal sentiment analysisrdquo in Proceedingsof the Conference on Empirical Methods in Natural LanguageProcessing EMNLP 2015 pp 2539ndash2544 prt September 2015

[26] Y Zheng ldquoAdapt to Emotional Reactions In Context-awarePersonalizationrdquo in Proceeding of the 4thWorkshop on Emotionsand Personality in Personalized Systems (EMPIRE) p 1 2016

[27] S M Mohammad and P D Turney ldquoCrowdsourcing a word-emotion association lexiconrdquo Computational Intelligence vol29 no 3 pp 436ndash465 2013

[28] Y H Hu P J Lee K Chen J M Tarn and D V DangldquoHotel Recommendation System based on Review and ContextInformation a Collaborative filtering Apprordquo in PACIS p 2212016

[29] The Macquarie Thesaurus Macquarie Library J Bernard EdSydney Australia 1986

[30] S Aman and S Szpakowicz ldquoIdentifying expressions of emo-tion in textrdquo in International Conference on Text Speech andDialogue pp 196ndash205 Berlin Germany 2007


[31] H Tang S Tan andXCheng ldquoA survey on sentiment detectionof reviewsrdquo Expert Systems with Applications vol 36 no 7 pp10760ndash10773 2009

[32] S Gong ldquoA collaborative filtering recommendation algorithmbased on user clustering and item clusteringrdquo Journal of Soft-ware vol 5 no 7 pp 745ndash752 2010

[33] X Su andTMKhoshgoftaar ldquoA survey of collaborative filteringtechniquesrdquo Advances in Artificial Intelligence vol 4 2009

[34] T Dunning ldquoAccurate methods for the statistics of surprise andcoincidencerdquo Computational Linguistics vol 19 no 1 pp 61ndash741993

[35] M-Y Hsieh W-K Chou and K-C Li ldquoBuilding a mobilemovie recommendation service by user rating and APP usagewith linked data on Hadooprdquo Multimedia Tools and Applica-tions vol 76 no 3 pp 3383ndash3401 2017

[36] M Papagelis and D Plexousakis ldquoQualitative analysis of user-based and item-based prediction algorithms for recommenda-tion agentsrdquo Engineering Applications of Artificial Intelligencevol 18 no 7 pp 781ndash789 2005

[37] S Schelter and S Owen ldquoCollaborative filtering with apachemahoutrdquo in Proceedings of the ACM RecSys Challenge 2012

[38] G Shani and A Gunawardana ldquoEvaluating recommendationsystemsrdquo Recommender Systems Handbook pp 257ndash297 2011

[39] M Taboada J Brooke M Tofiloski K Voll and M StedeldquoLexicon-based methods for sentiment analysisrdquo Computa-tional Linguistics vol 37 no 2 pp 267ndash307 2011

[40] E Cambria R Speer C Havasi and A Hussain ldquoSenticNetA publicly available semantic resource for opinion miningrdquoin Proceedings of the 2010 AAAI Fall Symposium pp 14ndash18November 2010

Submit your manuscripts athttpswwwhindawicom

Computer Games Technology

International Journal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014


Distributed Sensor Networks


Advances in

FuzzySystems

Hindawi Publishing Corporationhttpwwwhindawicom

Volume 2014


ReconfigurableComputing

Hindawi Publishing Corporation httpwwwhindawicom Volume 2014


Applied Computational Intelligence and Soft Computing

thinspAdvancesthinspinthinsp

Artificial Intelligence

HindawithinspPublishingthinspCorporationhttpwwwhindawicom Volumethinsp2014

Advances inSoftware EngineeringHindawi Publishing Corporationhttpwwwhindawicom Volume 2014


Electrical and Computer Engineering

Journal of

Hindawi Publishing Corporation

httpwwwhindawicom Volume 2014

Advances in

Multimedia


Biomedical Imaging


Advances in


RoboticsJournal of



Computational Intelligence and Neuroscience

Industrial EngineeringJournal of


Modelling amp Simulation in EngineeringHindawi Publishing Corporation httpwwwhindawicom Volume 2014

The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014


Human-ComputerInteraction

Advances in

Computer EngineeringAdvances in



Predictions

Recommendations active users

Items

UsersRatings table

+

TopmdashN list of items for

j for active userPajmdashprediction on item

I1 I2 I3 Ii ImU1

U2

Ua

Un

middot middot middotmiddot middot middot

Figure 1 Collaborative filtering process

IAPS and 1000 songs datasets [7] but none of these relate totourist destination recommendation data

In development of the proposed system we addressed thepaucity of datasets with contextual parameters by employingSentimental Analysis (SA) to acquire the emotional states ofusers based on user reviews for 100 destinations We thenderived emotional tags and manually compared the accuracyof the tags for each of the destinations

In this study a CARS that uses partial contextual userpreferences in the form of user item context and ratingwas developed This is unlike a traditional recommendationsystem which is based on knowledge of preferences of aset of items and the input data is in the form of user itemand ratingThe contextual information that we consider herecan be applied to various stages of the recommendationprocess A CAR process based on contextual user prefer-ences elicitation and estimation can take one of three formsas follows contextual prefiltering contextual postfilteringor contextual modeling In this study we implemented acontextual prefiltering paradigm-based solution and use thecontextual information to select themost relevant item times userdata for generating recommendations [8] The prefilteringapproach allows for the deployment of any of the numeroustraditional recommendation techniques previously proposed[9]

The main objectives of this study are as follows

(1) Deriving suitable datasets for tourist destination rec-ommendations considering emotion as a contextualparameter

(2) Examining how emotion influences recommendationof destinations using prefiltering techniques

2 Background

The body of literature related to recommendation systemsis constantly expanding Previous studies have introducedalgorithms for recommendations improved ways of buildinguser models to represent user preferences interests andbehaviors and domain-specific applications [10] Petrevskaand Koceski (2012) proposed a tourism recommendationsystem based on user preferences interests and desires andsuggested tourist attractions based on profiling user behav-iors such as reading other reviews tomake decisions [11] Des-tinations were observed as objects and changes in user pref-erences following subsequent visits tracked according to therating and user behavior was considered in the profiling and

recommendations generated based on collaborative filteringSarkaleh et al (2012) suggested a tourism recommendationsystem using location and user features as contextual param-eters [12] Zheng et al (2012) proposed a CARS for travel rec-ommendations using differential contextual relaxation andconsidered trip type duration originality destination cityand month as contextual parameters with user collaborativefiltering to propose a hybrid recommendation to mitigate thedisadvantages of the solo use of prefiltering and postfilteringtechniques [13] De Pessemier et al (2015) suggested a systemwith group recommendation for traveler destinations basedon a userrsquos rating profile personal interests and specificationfor their next destination by following a hybrid approachin combination with content-based collaborative filteringand knowledge-based strategies [14] Collaborative filteringis a widely adapted recommendation algorithm in whichpredictions and recommendations are based on the ratingsor the behavior of the users in the system [15 16]

The approach considers a list of n number of users 119880 andlist of 119898 number of items 119868 As shown in Figure 1 the user119906119886 119906119886 isin 119880 is called the active user for whom the task of thecollaborative filtering algorithm is to find an item likelinessthat can be in two forms 119875119886119895 is the predictive rating for item119894 for the active user 119880119886 where 119895 isin 119898 The prediction rating isa numerical value with the same scale as the rating providedby user 119906119886The recommendation is the list of119873 items that theactive user will like the most [17]

Collaborative filtering techniques can be either user-based or item-based The user-based collaborative filteringmodel recommends items based on computing similar neigh-bors and creates a group of users that are compatible with atarget user Item-based collaborative filtering computes sim-ilarity based on items and finds items that are similar to thegiven userrsquos rated items [18] A traditional recommendationsystem starts by estimating the rating function R using aninitialized set of ratings that are either explicitly provided byusers or are implicitly inferred by the system for user 119906 anditem 119894 that has not been rated

R user times item 997888rarr rating (1)In a CARS preferences are predicted by incorporating

available contextual information into the recommendationprocess and R is estimated by

R user times item times context 997888rarr rating (2)where user and item are the space of users and items andrating is the space of rating for the user and item pairs The



C U































Emotion tagcreation





Precision =119862

119860(3)





User



list


Database

Emotion tags


User times place











Similarity distance





sum119899isin119873 (sim (119906 119899) + 1) (4)



1 0 1 11 1 0 11 0 0 11 0 0 1


U1

U2

U3

U4


1 075 066 08075 1 05 06

1 04 1 051 05 1 075


U1

U2

U3

U4




11987511990611199041


(1 + 075 + 066 + 08)

=(2 + 0 + 166 + 18)

321= 17

11987511990611199042


(075 + 1 + 05 + 06)

=(175 + 2 + 0 + 16)

285= 188

(5)






Total Ns Nn N







precision =119873119903119904119873119904

(6)


Ave119875 =sum119899119896=1 (119875 (119896) times rel (119896))


MAP =sum119902119902=1 Ave119875 (119902)

119876

(7)




644946

552100

833

0 20 40 60 80 100 120CFN




Alg

orith

m




MAPE =sum119862119888=1sum

119902119902=1 Ave119875 (119902)

sum119862119888=1 119876 (8)







5 Conclusion













References


















































Advances in

FuzzySystems


Volume 2014












Journal of



Advances in

Multimedia


Biomedical Imaging


Advances in


RoboticsJournal of










Advances in





C U































Emotion tagcreation





Precision =119862

119860(3)





User



list


Database

Emotion tags


User times place











Similarity distance





sum119899isin119873 (sim (119906 119899) + 1) (4)



1 0 1 11 1 0 11 0 0 11 0 0 1


U1

U2

U3

U4


1 075 066 08075 1 05 06

1 04 1 051 05 1 075


U1

U2

U3

U4




11987511990611199041


(1 + 075 + 066 + 08)

=(2 + 0 + 166 + 18)

321= 17

11987511990611199042


(075 + 1 + 05 + 06)

=(175 + 2 + 0 + 16)

285= 188

(5)






Total Ns Nn N







precision =119873119903119904119873119904

(6)


Ave119875 =sum119899119896=1 (119875 (119896) times rel (119896))


MAP =sum119902119902=1 Ave119875 (119902)

119876

(7)




644946

552100

833

0 20 40 60 80 100 120CFN




Alg

orith

m




MAPE =sum119862119888=1sum

119902119902=1 Ave119875 (119902)

sum119862119888=1 119876 (8)







5 Conclusion













References


















































Advances in

FuzzySystems


Volume 2014












Journal of



Advances in

Multimedia


Biomedical Imaging


Advances in


RoboticsJournal of










Advances in























Emotion tagcreation





Precision =119862

119860(3)





User



list


Database

Emotion tags


User times place











Similarity distance





sum119899isin119873 (sim (119906 119899) + 1) (4)



1 0 1 11 1 0 11 0 0 11 0 0 1


U1

U2

U3

U4


1 075 066 08075 1 05 06

1 04 1 051 05 1 075


U1

U2

U3

U4




11987511990611199041


(1 + 075 + 066 + 08)

=(2 + 0 + 166 + 18)

321= 17

11987511990611199042


(075 + 1 + 05 + 06)

=(175 + 2 + 0 + 16)

285= 188

(5)






Total Ns Nn N







precision =119873119903119904119873119904

(6)


Ave119875 =sum119899119896=1 (119875 (119896) times rel (119896))


MAP =sum119902119902=1 Ave119875 (119902)

119876

(7)




644946

552100

833

0 20 40 60 80 100 120CFN




Alg

orith

m




MAPE =sum119862119888=1sum

119902119902=1 Ave119875 (119902)

sum119862119888=1 119876 (8)







5 Conclusion













References


















































Advances in

FuzzySystems


Volume 2014












Journal of



Advances in

Multimedia


Biomedical Imaging


Advances in


RoboticsJournal of










Advances in












Emotion tagcreation





Precision =119862

119860(3)





User



list


Database

Emotion tags


User times place











Similarity distance





sum119899isin119873 (sim (119906 119899) + 1) (4)



1 0 1 11 1 0 11 0 0 11 0 0 1


U1

U2

U3

U4


1 075 066 08075 1 05 06

1 04 1 051 05 1 075


U1

U2

U3

U4




11987511990611199041


(1 + 075 + 066 + 08)

=(2 + 0 + 166 + 18)

321= 17

11987511990611199042


(075 + 1 + 05 + 06)

=(175 + 2 + 0 + 16)

285= 188

(5)






Total Ns Nn N







precision =119873119903119904119873119904

(6)


Ave119875 =sum119899119896=1 (119875 (119896) times rel (119896))


MAP =sum119902119902=1 Ave119875 (119902)

119876

(7)




644946

552100

833

0 20 40 60 80 100 120CFN




Alg

orith

m




MAPE =sum119862119888=1sum

119902119902=1 Ave119875 (119902)

sum119862119888=1 119876 (8)







5 Conclusion













References


















































Advances in

FuzzySystems


Volume 2014












Journal of



Advances in

Multimedia


Biomedical Imaging


Advances in


RoboticsJournal of










Advances in





User



list


Database

Emotion tags


User times place











Similarity distance





sum119899isin119873 (sim (119906 119899) + 1) (4)



1 0 1 11 1 0 11 0 0 11 0 0 1


U1

U2

U3

U4


1 075 066 08075 1 05 06

1 04 1 051 05 1 075


U1

U2

U3

U4




11987511990611199041


(1 + 075 + 066 + 08)

=(2 + 0 + 166 + 18)

321= 17

11987511990611199042


(075 + 1 + 05 + 06)

=(175 + 2 + 0 + 16)

285= 188

(5)






Total Ns Nn N







precision =119873119903119904119873119904

(6)


Ave119875 =sum119899119896=1 (119875 (119896) times rel (119896))


MAP =sum119902119902=1 Ave119875 (119902)

119876

(7)




644946

552100

833

0 20 40 60 80 100 120CFN




Alg

orith

m




MAPE =sum119862119888=1sum

119902119902=1 Ave119875 (119902)

sum119862119888=1 119876 (8)







5 Conclusion













References


















































Advances in

FuzzySystems


Volume 2014












Journal of



Advances in

Multimedia


Biomedical Imaging


Advances in


RoboticsJournal of










Advances in




1 0 1 11 1 0 11 0 0 11 0 0 1


U1

U2

U3

U4


1 075 066 08075 1 05 06

1 04 1 051 05 1 075


U1

U2

U3

U4




11987511990611199041


(1 + 075 + 066 + 08)

=(2 + 0 + 166 + 18)

321= 17

11987511990611199042


(075 + 1 + 05 + 06)

=(175 + 2 + 0 + 16)

285= 188

(5)






Total Ns Nn N







precision =119873119903119904119873119904

(6)


Ave119875 =sum119899119896=1 (119875 (119896) times rel (119896))


MAP =sum119902119902=1 Ave119875 (119902)

119876

(7)




644946

552100

833

0 20 40 60 80 100 120CFN




Alg

orith

m




MAPE =sum119862119888=1sum

119902119902=1 Ave119875 (119902)

sum119862119888=1 119876 (8)







5 Conclusion













References


















































Advances in

FuzzySystems


Volume 2014












Journal of



Advances in

Multimedia


Biomedical Imaging


Advances in


RoboticsJournal of










Advances in






644946

552100

833

0 20 40 60 80 100 120CFN




Alg

orith

m




MAPE =sum119862119888=1sum

119902119902=1 Ave119875 (119902)

sum119862119888=1 119876 (8)







5 Conclusion













References


















































Advances in

FuzzySystems


Volume 2014












Journal of



Advances in

Multimedia


Biomedical Imaging


Advances in


RoboticsJournal of










Advances in









References


















































Advances in

FuzzySystems


Volume 2014












Journal of



Advances in

Multimedia


Biomedical Imaging


Advances in


RoboticsJournal of










Advances in





















Advances in

FuzzySystems


Volume 2014












Journal of



Advances in

Multimedia


Biomedical Imaging


Advances in


RoboticsJournal of










Advances in










Advances in

FuzzySystems


Volume 2014












Journal of



Advances in

Multimedia


Biomedical Imaging


Advances in


RoboticsJournal of










Advances in



Documents

The Prefiltering Techniques in Emotion Based Place …downloads.hindawi.com/journals/acisc/2017/5680398.pdf · 2019-07-30 · The Prefiltering Techniques in Emotion Based Place Recommendation