NoReC: The Norwegian Review Corpus

This is the official repository for the Norwegian Review Corpus (NoReC, ver. 2.1), created for the purpose of training and evaluating models for document-level sentiment analysis. More than 43,000 full-text reviews have been collected from major Norwegian news sources and cover a range of different domains, including literature, movies, video games, restaurants, music and theater, in addition to product reviews across a range of categories. Each review is labeled with a manually assigned score of 1–6, as provided by the rating of the original author. The accompanying paper by Velldal et al. at LREC 2018 describes the initial (ver. 1) release of the data in more detail.

Dataset details

• The columns are: id, split, rating, category, day, month, year, excerpt, language, source, authors, title, url, text where basic usage has text as the input and rating as the output.
• Curated by: NoReC was created as part of the SANT project (Sentiment Analysis for Norwegian Text), coordinated by the Language Technology Group (LTG) at the University of Oslo, in collaboration with the Norwegian Broadcasting Corporation (NRK), Schibsted Media Group and Aller Media.
• Funded by: The SANT project is funded by the Research Council of Norway (NFR grant number 270908).
• Shared by: The SANT project (Sentiment Analysis for Norwegian Text) at the Language Technology Group (LTG) at the University of Oslo
• Language(s) (NLP): Norwegian Bokmål (nb) and Norwegian Nynorsk (nn))
• License: The data is distributed under a Creative Commons Attribution-NonCommercial licence (CC BY-NC 4.0), access the full license text here: https://creativecommons.org/licenses/by-nc/4.0/ The licence is motivated by the need to block the possibility of third parties redistributing the orignal reviews for commercial purposes. Note that machine learned models, extracted lexicons, embeddings, and similar resources that are created on the basis of NoReC are not considered to contain the original data and so can be freely used also for commercial purposes despite the non-commercial condition.
• Dataset Sources This version of the corpus (v.2.1) comprises 43,436 review texts extracted from eight different news sources: Dagbladet, VG, Aftenposten, Bergens Tidende, Fædrelandsvennen, Stavanger Aftenblad, DinSide.no and P3.no.
• Repository: https://github.com/ltgoslo/norec
• Paper : The accompanying paper by Velldal et al. at LREC 2018 describes the (initial release of the) data in more detail.

Uses

The dataset is intended for document-level sentiment analysis, to learn to predict the rating from the text. The field category can be considered the "domain" of each text. By filtering in and out category values, one may inspect cross-domain performance of a model. The other fields contain metadata preserved from the original version of the dataset, and may be used for further filtering and analyses.

Source Data

This 2nd release, v.2.1 of the corpus comprises 43,436 review texts extracted from eight different news sources: Dagbladet (db), VG (vg), Aftenposten (ap), Bergens Tidende (bt), Fædrelandsvennen (fvn), Stavanger Aftenblad (sa), DinSide.no (dinside) and P3.no (p3).

In terms of publishing date the reviews mainly cover the time span 2003–2019, although it also includes a handful of reviews dating back as far as 1998.

Some statistics

Distribution over year and publication source

All splits combined

year	ap	bt	db	dinside	fvn	p3	sa	vg	Total
2003*	0	4	0	143	0	25	0	286	458
2004	0	44	0	142	0	12	19	984	1201
2005	0	0	0	179	0	6	224	909	1318
2006	0	0	0	240	0	11	294	778	1323
2007	0	0	0	139	0	127	400	725	1391
2008	0	0	0	119	0	216	369	739	1443
2009	0	52	377	163	27	428	259	815	2121
2010	0	100	642	260	156	571	309	769	2807
2011	1	51	592	284	146	652	362	900	2988
2012	2	150	613	257	332	611	561	763	3289
2013	4	160	527	216	213	619	433	1058	3230
2014	39	291	501	236	357	546	387	1191	3548
2015	249	235	728	245	456	499	620	849	3881
2016	309	340	809	177	321	439	682	715	3792
2017	649	491	921	248	692	567	822	687	5077
2018	605	470	885	194	466	339	860	492	4311
2019	260	167	95	30	160	36	346	165	1259

2003*: Including the 31 documents 1998-2002

Distribution over split and rating

split	1	2	3	4	5	6	Total
dev	51	225	707	1409	1678	278	4348
test	27	242	706	1385	1714	266	4340
train	379	2287	6004	11304	12614	2161	34749

Distribution over split and category

split	games	literature	misc	music	products	restaurants	screen	sports	stage	Total
dev	179	539	28	1445	347	94	1569	15	132	4348
test	180	547	24	1444	345	98	1579	16	107	4340
train	1453	4337	156	11777	2771	745	12536	118	856	34749

Citation

@InProceedings{VelOvrBer18,
  author = {Erik Velldal and Lilja {\O}vrelid and Eivind Alexander Bergem and  Cathrine Stadsnes and Samia Touileb and Fredrik J{\o}rgensen},
  title = {{NoReC}: The {N}orwegian {R}eview {C}orpus},
  booktitle = {Proceedings of the 11th edition of the 
               Language Resources and Evaluation Conference},
  year = {2018},
  address = {Miyazaki, Japan},
  pages = {4186--4191}
}

Dataset Card Authors

Vladislav Mikhailov and Erik Velldal

Dataset Card Contact

vladism@ifi.uio.no and erikve@ifi.uio.no