epinions — Epinions review ratings and trust (2001-2003)

Description

The "extended Epinions dataset", which the staff of the consumer review site Epinions.com gave to Paolo Massa, and which he distributed on the TrustLet wiki: the helpfulness ratings that the users gave to the reviews and essays (articles) written by other users, and the trust and distrust statements among the users, including the distrust statements, which the site did not show publicly. Users and articles are identified by anonymized numeric ids. The files are obtained from a copy of the dataset on Kaggle (the TrustLet site no longer exists).

ratings: a bipartite network of the users and the articles they rated. Nodes are the users who rated some article, and the rated articles (users and articles have separate id spaces), and an edge connects a user to an article they rated. The rated objects are articles, not products: each rating states how helpful a user found an article of another user, on a scale of 1 ("not helpful") to 5 ("most helpful"). The data contain also 6 as rating value (in about 0.7% of the ratings), which the original documentation does not mention (the Kaggle copy says to treat it as 5); the values are kept as they are. The creation dates of the ratings range from 10 January 2001 to May 2002, but about 90% of the ratings bear the date of 10 January 2001, which is apparently the earliest date recorded (Epinions was founded in 1999), so that the dates of the earlier ratings are not known. The file of the ratings ends in the middle of its last line, without the category of that rating (and possibly cut short at the source: the copies of the file on TrustLet and on Kaggle are the same); that rating is kept, with the category -1. The authors and subjects of the articles are given for the articles in the file mc.txt of the dataset.
trust: a directed signed network of the trust and distrust statements among the users (717,667 trusts and 123,705 distrusts), with their dates, from January 2001 to August 2003. Nodes are the users making or receiving some statement (in the same id space as the users of ratings and the authors of the articles), and a directed edge from user A to user B is a statement of A about B (573 users made a statement about themselves). These are the same data as the network epinions of the entry signed_social_media (from SNAP, with the same numbers of statements and of users, 131,828), here with the original user ids and the dates.

The networks have the following properties:

(Node) is_user (ratings): whether the node is a user (otherwise an article).
(Node) id (ratings): the anonymized id of the user or article.
(Node) author_id (ratings): the id of the user who wrote the article (-1 for users, and for the articles not listed in mc.txt).
(Node) subject_id (ratings): the id of the subject the article is about (-1 for users, for the articles not listed in mc.txt, and for the 341 articles of mc.txt without a subject).
(Edge) rating (ratings): the rating value (1 to 5, or the undocumented value 6).
(Edge) status (ratings): the display status of the rating (1 if the user chose not to show their name next to the rating, 0 otherwise).
(Edge) time (ratings): the date the article was first rated by the user (Unix time, at 00:00 UTC of that date).
(Edge) last_modified (ratings): the date the rating was last modified (Unix time, at 00:00 UTC; -1 if not given).
(Edge) type (ratings): the type of the rated object (1 or 2; the meaning of the values is not documented).
(Edge) vertical_id (ratings): the id of the category ("vertical") of the article (-1 for about 7% of the ratings, which the documentation does not explain, and for the incomplete last rating).
(Node) id (trust): the anonymized id of the user.
(Edge) sign (trust): +1 for trust and -1 for distrust.
(Edge) time (trust): the date of the statement (Unix time, at 00:00 UTC of that date).

Tags

Economic Preferences Social Online Bipartite Weighted Signed Timestamps Metadata

Added

2020-08-11

Citation

P. Massa, P. Avesani, "Controversial users demand local trust metrics: An experimental study on Epinions.com community." Proceedings of the 20th National Conference on Artificial Intelligence (AAAI-05), 121-126 (2005) cdn.aaai.org
M. R. Hamedani, I. Ali, J. Hong, S.-W. Kim, "TrustRec: An effective approach to exploit implicit trust and distrust relationships along with explicit ones for accurate recommendations." Computer Science and Information Systems 18(1), 93-114 (2021) 10.2298/csis200608039h Sci-Hub

Upstream URL OK

https://web.archive.org/web/20080514161127/http://www.trustlet.org/wiki/Extended_Epinions_dataset
https://www.kaggle.com/datasets/masoud3/epinions-trust-network

Upstream license

Other (no license stated; Kaggle dataset masoud3/epinions-trust-network), https://www.kaggle.com/datasets/masoud3/epinions-trust-network

Networks

Tip: click on a column header to sort the list, or hover over it for a description.

Column legend

$\left<k\right>$: Mean degree
$\sigma_k$: Standard deviation of the degrees
$\lambda_h$: Largest eigenvalue of the Hashimoto (non-backtracking) matrix
$\tau$: Random walk mixing time
$r$: Degree assortativity
$c$: Global clustering coefficient
$\oslash$: (Pseudo-) diameter
$S$: Relative size of the largest component
NPs: Node properties
EPs: Edge properties

Name	Nodes	Edges	$\left<k\right>$	$\sigma_k$	$\lambda_h$	$\tau$	$r$	$c$	$\oslash$	$S$	Kind	Mode	NPs	EPs	gt	GraphML	GML	csv
ratings	876,252	13,668,320	31.20	396.37	838.83	45.13	-0.12	0.00	12	1.00	Undirected	Bipartite	is_user id author_id subject_id	rating status time last_modified type vertical_id	43.8 MiB	101.1 MiB	94.8 MiB	96.8 MiB
trust	131,828	841,372	6.38	60.55	240.57	119.18	-0.04	0.11	16	0.90	Directed	Unipartite	id	sign time	4.3 MiB	7.5 MiB	7.2 MiB	6.4 MiB