This dataset contains a collection of interaction sequences between allies in online Diplomacy [1] games. A sequence consists of consecutive game seasons during which the two players exchange messages and help each other in the game. Half of the sequences end with betrayal, while the other half are part of lasting friendships. Description Diplomacy [1] is a popular and engaging strategic board game that is often played online [2, 3]. It is based heavily on communication between the players. Due to its military domination setting, Diplomacy is a well suited environment for studying naturally occurring betrayal and deception. From a collection of Diplomacy game logs, we identified and extracted *ongoing, established, and reciprocal* friendships: relationships that contain at least two consecutive and reciprocated acts of support that span at least three seasons in game time, with no more than five seasons passing between two acts of friendship. We then identified 250 *betrayals*: the subset of friendships described above that are followed by at least two attacks. To match each betrayal, we selected a friendship that is not followed by any offensive action, but is otherwise nearly identical (in terms of length and relative time within the game). The current dataset consists of these selected betrayals and friendships only. Each relationship contains a sequence of seasons. Within each season, we provide features extracted from the messages sent by each player. Acknowledgements: This dataset is distributed under the [Open Data Commons Attribution (ODC-By 1.0)](http://opendatacommons.org/licenses/by/summary/) license. It was collected by Vlad Niculae, Srijan Kumar, Jordan Boyd-Graber and Cristian Danescu-Niculescu-Mizil. If you use this dataset, please cite this paper: > Niculae, V., Kumar, S., Boyd-Graber, J., & Danescu-Niculescu-Mizil, C. (2015). Linguistic harbingers of betrayal: A case study on an online strategy game. In Proceedings of the ACL. Data format The dataset is a UTF-8 encoded JSON file, which can be loaded into a Python kernel with the following code: >>> import json >>> from io import open >>> with open("diplomacy_data.json