Explore our website to discover the perfect beer that aligns with your mood: Datastory of The Beer Mood Matcher
Every beer tells a story, evoking a spectrum of emotions in its drinker. By delving into the rich textual content of the reviews, we can find the perfect beer, matches our desired mood. Our research is focused on determining how different beers relate to various moods expressed in reviews on popular platforms of BeerAdvocate and RateBeer. Reviews are analyzed and scored based on primary emotions including joy, sadness, anger, fear, love, and surprise, applying NLP techniques, specifically Sentence-Bert for semantic similarity detection. Furthermore, we have determined the relationship between the characteristics of various beers and the moods they evoke, as identified in user reviews, which can benefit the marketing industry. Moreover, the connections between the emotions users exhibit while writing reviews and their various ratings are examined. Finally, a mood-based recommendation list of beer styles is created based on the mapped emotion scores.
- How do the characteristics of different beers align with the moods identified in reviews?
- In what moods do users tend to write comments?
- Does a specific correlation exist between beer styles and the range of emotions identified in consumer reviews?
- Is there a pattern between higher-rated beers and reviews scored higher by positive emotions ?
- What kinds of moods do the most-popular beers tend to match?
- Are there any regional distribution characteristics of users’ moods towards beer?
- What is the perfect beer for any mood?
The primary emotions listed in Figure 1 below is used to score the reviews. It is selected by analyzing the literature of emotion detection, and these emotions are found to be the main headings of emotions known as Parrot's emotion model.
Figure 1: Parrot’s emotions model 1
After applying text-based emotion detection, we will use an emotion map to obtain the moods of the beer styles2
- Working on two different datasets of different websites: BeerAdvocate and RateBeer
- Excluding beers with a minimal number of reviews to have a more represantative data.
- Descriptive statistics: List of features and their quantity
- Utilizing charts to enhance understanding of the data.
- Analysis and visualization of the review dataset
- Stemming, lemmitization, stopword removal (NLTK library)
- Calculating wordcount of each review and each beer
- Visualizing the most frequent n-grams (n = 1, 2, 3)
-
Inputs are review paragraphs, and similarity scores is calculated based on a designated word list for each emotion, using the SentenceTransformers framework within an SBert model 3.
- By averaging the scores from reviews for the same beer, we have generated a list of average scores across six emotions for each type of beer.
- A vector of 6 emotions are obrained by averaging the scores of all beers under the same style.
Based on high and low emotions of each beer style, assigning the most representative mood to each beer style is assigned.
Based on the emotion scores, the relationship between mood and other potential variables are further analyzed, for example, there might be some regional distribution patterns of user’s mood towards beer. Besides, we have related mood to beer characteristics and test the correlation of them, to explore whether beer’s appearance, aroma, or taste would influence drinker’s mood.
-
Regional distribution of reviews’ emotions
- Creation of a geographical heat map or histogram to visualize the emotion scores for different regions
- Performing statistical tests, such as t-tests or ANOVA, to compare whether there are significant differences in emotion scores across regions.
-
Correlation between emotion and beer’s characteristics (appearance, aroma, taste)
- Creating a scatter plot to show the relationship between different beer characteristic scores and emotion scores.
- Using correlation analysis such as Pearson's correlation coefficient to quantify the linear relationship between them. If there is a linear correlation, use linear regression to fit the model.
For different beers, there could be interesting study on the potential relationship between emotions and some beer attributes.
As some examples; normally, users are expected to review a beer with positive emotions when they give high ratings, so we will explore the correlation between emotion and beer ratings, then try to discover the “top-rated moods”. By applying the methods explained in 4.1. we can calculate correlation between emotions and beer ratings and discover the emotions corresponding to the top-rated beers, then there could be some marketing strategy for brewers, for example, creating an atmosphere with a related “top-rated mood” when marketing the beer.
Moreover, drunkenness usually leads to mood swings, so do we get anger after getting high, or feel really relaxed? We will study the correlation between emotions and ABV to find the answer.
- The negative emotions will be eliminated for the final recommandation list since we would not want to recommend any beer in those moods.
- We will provide a list of beer styles and a their moods matched according to the methodology above of mapped emotions.
| Team Member | Roles |
|---|---|
| Yiwei | NLP + Webpage |
| Yunlong | Analysis of the Results + Webpage |
| Victor | Mood Map + Data Story |
| Zimu | Analysis of the Results + Data Story |
| Nil | NLP + Mood Map + Data Story |
Footnotes
-
S. Zad, M. Heidari, J. H. J. Jones and O. Uzuner, "Emotion Detection of Textual Data: An Interdisciplinary Survey," 2021 IEEE World AI IoT Congress (AIIoT), Seattle, WA, USA, 2021, pp. 0255-0261, doi: 10.1109/AIIoT52608.2021.9454192. https://ieeexplore.ieee.org/abstract/document/9454192 ↩
-
E. Cambria, A. Livingstone and A. Hussain, "The hourglass of emotions" in Cognitive behavioural systems, Springer, pp. 144-157, 2012. ↩
-
Reimers, N., & Gurevych, I. (2019). Sentence-bert: Sentence embeddings using siamese bert-networks. arXiv preprint arXiv:1908.10084. https://www.sbert.net/ ↩