We solicit submission of papers of two to eight pages, representing reports of original research, preliminary research results, proposals for new work, descriptions of neural network based toolkits tailored for IR, and position papers. Papers presented at the Neu-IR 2017 will be required to be uploaded to arXiv.org but will be considered non-archival, and may be submitted elsewhere (modified or not), although the workshop site will maintain a link to the arXiv versions. This makes the workshop a forum for the presentation and discussion of current work, without preventing the work from being published elsewhere.
Neu-IR 2017 will have two tracks: a special track on resources and a general track.
Neu-IR 2017 will have two tracks: a special track on resources and a general track.
Special track on resources
For Neu-IR 2017 we are especially interested in tackling challenges around training, evaluation and the reproducibility of deep neural network models for IR. Most available IR datasets are of insufficient scale for training these large data hungry models. The rapid increase in the number of new models in the literature also puts significant burden on future authors with respect to the implementation of baselines. The fact that most of these existing models in the literature have been evaluated on different tasks–and even private industry datasets–compounds the issue. Finally, more should be done in terms of encouraging authors to share their model implementations and document exact hyper-parameters for reproducibility.
The Neu-IR community can significantly benefit from the existing of a public benchmark and a model repository. Towards this goal we encourage the community to submit proposals for,
The Neu-IR community can significantly benefit from the existing of a public benchmark and a model repository. Towards this goal we encourage the community to submit proposals for,
- New large scale benchmark collections appropriate for training and evaluating deep neural network models with millions (or billions) of parameters.
- Building a central shared model repository without enforcing the use of any specific NN toolkit.
- Making appropriate hardware resources (e.g., GPUs) available for academic research.
- New tools and bindings to enable smooth interfacing between traditional IR frameworks and recent neural network toolkits.
- Standardizing frameworks appropriate for evaluating deep neural network models.
- Automatic and semi-automatic methods for generating training material at scale.
General track
In addition, Neu-IR 2017 welcomes submissions relevant to the following main themes:
- The application of neural network models in IR tasks, including but not limited to:
- Full text document retrieval, passage retrieval, question answering
- Web search, searching social media, distributed information retrieval, entity ranking
- Learning to rank combined with neural network based representation learning
- User and task modelling, personalized search, diversity
- Query formulation assistance, query recommendation, conversational search
- Multimedia retrieval
- Fundamental modelling challenges faced in such applications, including but not limited to:
- Learning dense representations for long documents
- Dealing with rare queries and rare words
- Modelling text at different granularities (character, word, passage, document)
- Compositionality of vector representations
- Jointly modelling queries, documents, entities and other structured/knowledge data
- Best practices for research and development in the area, dealing with concerns such as:
- Finding sufficient publicly-available training data
- Baselines, test data, avoiding overfitting
- Neural network toolkits
- Real-world use cases, deployment at scale
Important
All papers will be peer reviewed (single-blind) by the program committee and judged by their relevance to the workshop, either to the special topic or to the general themes identified above, and their potential to generate discussion. All submissions must be formatted according to the latest ACM SIG proceedings template. Papers can be two (2) to eight (8) pages long. When submitting, authors have to submit their paper to one of two tracks: the "Special track on resources" or the "General track". At least one of the authors of each accepted paper must register for the workshop and present the paper in-person.
Submission
Submission url: https://easychair.org/conferences/?conf=neuir2017
Important dates
- March 17, 2017: first call for papers
- April 12, 2017: second call for papers
- June 1, 2017: final call for papers
- June 25, 2017: submission deadline
- July 14, 2017: notification
- July 25, 2017: camera ready version of accepted papers due
- August 11, 2017: Neu-IR 2017 workshop
