The Beer Mood Matcher

Firstly, Let's get an overview of the dataset!

We have two datasets of beer reviews, details as below:

RateBeer's cloud highlights "taste" and subtle characteristics like "bit," with additional focus on aroma, texture, and aftertaste through words like "nose," "mouthfeel," and "finish." The BeerAdvocate word cloud emphasizes "head," "malt," and "taste," focusing on foam quality and flavour, while "body" and color terms are less frequent.

The distribution of the Emotion Scores

The chart shows that the emotion scores for the two websites under analysis are distributed between -0.2 and 0.8. The feelings of surprise and joy have higher average scores, indicating that these are prevalent emotions in beer reviews. On the other hand, fear has the lowest average, suggesting that it is not as frequently connected to the experience of analysing beer. 

The low standard deviations found in all of the emotion scores point to consistency in emotional reactions, providing the reliability of our data in expressing the emotions associated with each type of beer. This consistency makes it possible for us to apply these findings to individual beer experiences and to confidently link particular emotional profiles with particular beer styles.

Firstly, it is possible to say that the correlation between the emotions in comments on the RateBeer site is less compared to the BeerAdvocate site. The lowest similarity between joy and fear emotions on the RateBeer website, coming in at 0.22, indicates that our model distinguishes well between comments that contain emotions of fear and those of joy. The second lowest correlation score, with 0.29, is between joy and anger emotions similar to fear, suggesting that there are some distinctions for some of the opposite emotions on the RB website.

On the other hand, on both websites, the correlation between the sadness and love emotion scores is the most notable. At this point, perhaps we might question the melancholy behind love (!?) and this might indicate a challenge in distinguishing these emotions or an interesting psychological linkage between them. Also, it's possible to say that our model doesn't distinguish these two emotions very well.

When we interpret this overall scenario, it's possible to say that for both websites, the emotion of joy is more differentiated from the other emotions and seems to be more distinctly expressed or identified in the reviews compared to other emotions across both websites.

❔Is there a pattern between higher-rated beer characteristics and reviews scored higher by positive emotions ?

It aligns with our expectations that positive emotions, such as joy_score and love_score, demonstrate a positive correlation with beer characteristics, while most negative emotions exhibit negative correlations with these traits. In both the RB and BA datasets, we observe that love and joy scores show the highest positive correlations with these characteristics. This correlation seems reasonable as individuals tend to prefer beers with superior taste and aroma.

However, we also notice that the correlation between all emotion scores and appearance is not as straightforward. Considering the disparities between appearance and other beer characteristics, we hypothesise that this discrepancy might arise because when individuals comment on a beer, they often emphasise taste-related aspects rather than its visual appeal. 

❔How does ABV in beverages correlate with different emotions in user reviews?

In both the RB and BA datasets, the anger_score and fear_score exhibit the highest absolute correlations with beer ABV, and notably, these are the only high positive correlations observed. The intriguing similarity in the structures of both graphs is obvious, particularly the prominence of the anger_score. This observation could be attributed to the fact that higher ABV correlates with a higher alcohol percentage, and stronger alcohol content has the potential to evoke emotions, particularly making individuals more prone to feelings of anger. 

❔Is there a regional distribution characterizing users' emotions towards beer?

❔How emotion scores distribute across beer styles?

Once we've calculated the average emotion scores for each type of beer, we observe how the scores are distributed across various styles of beer. 'Joy' emotion has the highest average score found in the RateBeer data sets, where averaged scores range from 0 to 0.25 for all emotions. Beer styles typically score lower on feelings like "sadness" and "fear." It's also possible to observe correlated emotions "love" and "anger" distributed between other emotions.

For every mood considered, it's representation in the emotions space is defined in what we called, the ‘Mood Map’.

Through these steps, we have created a Mood Map that tries to resonate with the complexity and subtlety of human emotions and moods. Our analysis not only provides a unique perspective on beer preferences but also enhances the beer selection experience, allowing individuals to choose beers that align with their current mood or the ambience they wish to create.

The mood mapping results from both websites were meticulously combined, resulting in a consolidated list of beer styles categorised by mood, irrespective of the website they originated from. Out of a total of 169 beer styles analysed across these platforms, 74 were successfully mapped to specific moods. This translates to a mood mapping success rate of about 44%.

To effectively communicate the findings of our analysis, we have developed visual representations. These visuals succinctly illustrate the alignment of various beer styles with distinct moods. Focusing initially on the beer styles that have been mapped to particular moods, here is the distribution of these moods across the successfully categorised styles : 

In our pie charts, each slice represents a mood and is weighted according to the popularity of the beer styles it encompasses. Here, popularity is gauged by the volume of reviews each style has received. A notable portion of beer styles is labelled as ‘Undetermined’ due to the limitations in our mood mapping process, which didn’t accommodate a mood categorization for every style. Nonetheless, it’s interesting to observe that the ‘Energetic’ mood stands out in both websites, reflecting a high preference for beer styles associated with this mood. In contrast, the 'Contemplative' mood has the smallest representation, indicating lesser popularity for beer styles that evoke a contemplative state. 

Don't forget to check out our table to select a beer style that suits your current mood and check to see if your favourite pub has it available 🍺!

Even though we're not beer experts, our list reveals a clustering of beer styles named with similar descriptive words. It appears that beer styles signified by the word "black" are associated with the mood of nostalgia, while beer styles denoted by the word "fruit" are matches with cheerful and adventurous moods. Interestingly, all stout beers match contemplative and nostalgic moods, showcasing how certain beer types can evoke specific emotional landscapes.

References

1. 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. 

2. E. Cambria, A. Livingstone and A. Hussain, "The hourglass of emotions" in Cognitive behavioural systems, Springer, pp. 144-157, 2012.

3. Reimers, N., & Gurevych, I. (2019). Sentence-bert: Sentence embeddings using siamese bert-networks. arXiv preprint arXiv:1908.10084. 

4. J. McAuley, J. Leskovec and D. Jurafsky, "Learning Attitudes and Attributes from Multi-aspect Reviews," 2012 IEEE 12th International Conference on Data Mining, Brussels, Belgium, 2012, pp. 1020-1025, doi: 10.1109/ICDM.2012.110.

5. Gael Lederrey and Robert West. 2018. When Sheep Shop: Measuring Herding Effects in Product Ratings with Natural Experiments. In Proceedings of the 2018 World Wide Web Conference (WWW '18). International World Wide Web Conferences Steering Committee, Republic and Canton of Geneva, CHE, 793–802.