User research through codeLanguage analysis of Amazon reviews of fitness trackers
Coding for Humanities | Prof Ishizaki Fall 2016Shruti AdityaChowdhury
Proposal 1
Fitness tracker manufacturers
Technology magazines
Academic journals User reviews
Analyse the use of language across different source• Number of words (complexity of argument)• Sentiment analyses to find variance within a source• Compare the nature of the content (no clue of what or how)
Proposal 2
• What features people think are important• How attitudes change over time• Possibly creating customer segments
User reviews Sentiment analysis
Appraisal framework
(Martin & White)
Pivot
Using the Attitude aspects of the Appraisal framework• Manually categorize words as affect, judgement or appreciation (create a dictionary)• Find words that correspond to the sub-categories and study how they are used in
reviews• See if there are co-relations between the use of ‘Attitude’ words and number of stars
Appraisal framework(Martin and
White)
Engagement
GraduationAttitude Judgement
Affect
Appreciation
Pivot
Categorizing words was time consuming and not helpful – needed context to understand
Bottomline• It’s very difficult to do since it involves analyses of text by word, sentence and
paragraph.
Version 1
Helping people find the ‘right’ fitness tracker for them – based on what people have written • Create a database of keywords - features and synonyms of the features• Parse reviews for keywords• Extract sentences• Analyze sentiment• Normalize the score with the star rating• Ask for user input and return the best option
User reviewsSentiment
analysis per product, review,
line, featureTop 7 products
Master features list
Related key words
Version 2
User reviews
959 pages
Sentiment analysis per
product,review and line for words related to
Master features list
Top 7 products Related key words
FeaturesWords related to behaviour
changeBehavior change
Details of words related to changeCode name List of wordsstate of mindintent intent,intention,intended,intending,mean, meant,meaning,hope,hoped,hopinghealth health,healthy,fit,fitness,strength,strong,stronger,weighthabit habit,routine,practice,practiselifestyle lifestyle,regimebuy buy,bought,purchased,gotbecause cos,because,since,therefore,hencemotivate motivate,motivates,motivated,motivating,motivationalaware aware,awareness,notice,noticed,concious,conciousness
changetime since,earlier,now,used toamount more,less,increase,decrease,increased,decreased,increasing,decreasing,samechange change,changing,changed,turned,transformed,transformation,improve,improved,improving,worse,worsen,worseningnotice notice,noticed,noticing
usegoal goal,aim,target,aimed,targetted,objectiveactivity yoga,tai-chi,meditation,aerobic,weight,weights,lift,lifting,lungebehavior behavior,behaviour,behaviors,behaviours
relationshipencourage encourage,encouragement,encouraging,encouraged,encouragementshelp help,helps,helped,helping
emotions happy,sad,angry,guilty,unhappy
Overview of code
1. Creating Sql databases instead of files – so that the data can be stored and analyzed (through queries) in multiple ways
2. Define functions for1. Finding the number of pages per product2. Getting different products’ reviews3. Getting all reviews per star rating4. Splitting the reviews into sentences5. Doing sentiment analysis6. Reading the database table for features and product7. Reading the database table for features and line 8. Populating the database tables
3. Running the code and populating the database
Findings
• Positive and negative reviewsPositive – change and motivation. Negative – features (inaccuracy), customer service, product design, app
• Time periodPeople use pedometers longer than fitness trackers – a year v/s 3 monthsKeywords of ‘hours, days, weeks, months’ was found regarding fitness trackers and ‘months’, ‘years’ with pedometers
• Language usedMuch more extreme than with one on one interviews – lack of judgement? Altruistic motive?
• CompetitionCompetition was perceived as positive
• DisplayPeople like seeing a detailed display with graphics and numbers rather than a general sense of activity
• GiftPeople buy them for others and in pairs
• Supplement to recoveryPeople use it to recover from illness – chemotherapy, knee replacementUsage among older people is high
• Emotion scaleVery happy > Happy > Sad > Not happy > Angry
Reflections
• Key wordsBalance between the right word (“Change”) and too many variations
• Quantitative data is not as helpful for design research “It motivates me to walk more.” x 50 times
• On Amazon – Have to rate before you write a review which may impact how people write
Reflections
• Limitations of sentiment analysis - “I've used it on my runs, lifting, hikes” gets a neutral score when it should be positive
• Helped me discover new features that were important to people but not mentioned on the websites – flower and water tracking
• A lot of first-hand data but still biased – how and why people write it, how I wrote the key words and how I did the qualitative analysis
Points of investigation – Features
CaloriesSteps TimeDistance Sleep Silent alarm
Multiple sports
Goals ActivitiesCardio Backlight CallText
Waterproof RemindersActivity switch LEDCalendar Timer
Top Related