Impact of Polish wine reviews on the customer’s decision making

Ewa Gwiazdecka, Michał Marcińczuk

Changing lifestyles and opening up to the world in the last thirty years in Poland has an impact on the structure of the alcohol sale. Wine, relatively more expensive than traditional spirits, is associated with the Western lifestyles. According to The International Wine & Spirit Research in 2016 Poles bought 90.7 million liters of still wine. The forecasts of of IWST for this year will be 97.1 million, which is about 27.5 % more than 5 years ago! Although the quantity of wine bought in Poland is increasing every year, the consumption per capita is extremely low with only 3,5-4,5 l. The fact that the wine sales are blooming does not mean that we know what we are drinking: an average Pole buy wine because s/he likes it or because someone has already bought it. The price is also very important. On the other hand, there is a variety of the wine reviews, blogs, but the complicated terminology of the domain plus the excessive metaphors used by wine writes are rather not very helpful for the untutored consumer.

In this work we describe the results of the automatic extraction and classification of opinions about wines and their aspects and discuss the method of assigning the sentiment to the extracted adjectives.

We have collected more than 8600 wine reviews from 20 different blogs written between 2009 and 2017. We obtained a good diversification of writing styles as the reviews are written by 100 different authors. The average number of texts per author is 86. The texts were pre-processed using the morphological tagger WCRFT (Radziszewski, 2013) (from the CLARIN-PL infrastructure[1]) and the shallow parser Spejd (Przepiórkowski, 2008). Then we applied a set of manually crafted rules to detect phrases containing opinions. Each opinion should consists of three elements: an aspect (a noun), a phrase containing an opinion (in most cases an adjective but also verb phrases) and a modifier (negation or adverb, e.g., very, strongly, extremely), for example: kwasowość ‘acidity’ [aspect] zbyt ‘too’ [adverbial modifier] agresywna ‘aggressive’ [opinion adjective]. The 15 top nouns which are used to describe wine are: kwasowość ‘acidity’, cena ‘price’, owoc ‘fruit’, zapach ‘smell’, smak ‘tase’, producent ‘producer’, tanina ‘tannin’, aromat ‘aromat’, butelka ‘bottle’, nuta ‘tone’, finisz ‘finish’, alkohol ‘alcohol’, kolor ‘color’, szczep ‘grape’, charakter ‘character’.

However, in such a specialized domain, the consumer may not be sure if the opinion associated with the aspect is positive, negative or “objective”. The use of so called “positive” adjective can be misleading like in delicate tannin which denotes a wine property.

We argue in this work that in order to determine the sentiment of the opinion phrase, we need to go beyond the positive-negative polarity and find a way to classify adjectives according to their semantic properties, but also taking into account the compositionality with the noun (Fahrni, Klenner 2008).

[1] http://clarin-pl.eu

References

Angela Fahrni and Manfred Klenner. 2008. Old wine or warm beer: Target-specific sentiment analysis of adjectives. In Proceedings of the Symposium on Affective Language in Human and Machine, AISB.

Adam Radziszewski, 2013. A tiered CRF tagger for Polish. In: Intelligent Tools for Building a Scientific Information Platform: Advanced Architectures and Solutions, ed(s). R. Bembenik, Ł.. Skonieczny, H. Rybiński, M. Kryszkiewicz, M. Niezgódka. Springer Berlin Heidelberg, pp. 215-230.

Adam Przepiórkowski. (2008). Powierzchniowe przetwarzanie języka polskiego. Warszawa: Akademicka Oficyna Wydawnicza EXIT.

Magdalena Zawisławska, Funkcja metafory w rekonstrukcji językowego obrazu świata na przykładzie metaforyki w języku winiarzy, „Poradnik Językowy” 1/2015, 79-88.

 

mots clés : vins ; polonais ; sentiment analysis ; natural language processing.

                                

Ewa Gwiazdecka, linguist and entrepreneur, a president of the management board in ASPEKTY Ldt. She received her Ph.D from Paris 4 – Srobonne in Mathematics, Logic, Computer Science applied to Human Sciences. She has taught linguistics and French grammar at Warsaw University. She founded Aspekty Ldt, where she develops tools for text mining and sentiment analysis.

Her research interest focus on lexical semantics, computational linguistics, tense, aspect and modality in languages.

  1. Ewa Gwiazdecka (2013) “Les verbes de mouvement – quelle modelisation pour le TALN?” in Des mots et du texte aux conceptions de la description linguistique, Wydawnictwa UW, 212-222.
  2. Bogacki K., Gwiazdecka E (2013) “Derivational structure of Polish verbs and the expansion of the dictionary”, in Automatic Processinf of Various Levels of Linguistic Phenomena, red. K. Vuckowić, B. Bekevac, M. Silberztein, Cambridge Scholars Publishing, 50-63

 

Michał Marcińczuk is an assistant professor in Department of Computational Intelligence at Wrocław University of Science and Technology. In 2012 Michał obtained a Ph.D. degree in computer science with a dissertation on information extraction from texts. Currently he is involved in several projects in the field of computational linguistics and natural language processing, including automatic recognition of named entities, temporal and spatial expressions, event description, sentiment analysis, plagiarism detection, lemmatization of multi-word expressions.

  1. Michał Marcińczuk, Marcin Oleksy and Jan Wieczorek (2016) Towards Recognition of Spatial Relations between Entities for Polish. In: Cognitive Studies | Études cognitives. Vol. 16, pp. 119-132
  2. Michał Marcińczuk (2015) Automatic construction of complex features in Conditional Random Fields for named entities recognition. In Angelova, G., Bontcheva, K. & Mitkov, R. (editors), Proceedings of International Conference Recent Advances in Natural Language Processing, pages 413–419. INCOMA Ltd.