Semantic Context Similarity

From Knoesis wiki
Jump to: navigation, search

Problem Description

Context is any information that can be used to characterize the situation of an entity. An entity is a person, place, or object that is considered relevant to the interaction between a user and an application, including the user and applications themselves as defined by [7][8]. Context has been an important aspects on many application areas such as Recommendation and Ambient Intelligence in comparing the similarity of two entities. TODO: Why is it important to compare two context?

In this work, we plan to develop a measure of semantic relatedness between two or more contexts that represent some state-of-the-world. Context may involve information of various types, including spatial-temporal information such as location, date and time, physical information such as traffic and weather or social information such as the people in a meeting or building, etc.

Notes

Currently we have identified two possible ways to approach the problem,

  • Represent context as a multidimensional model and use semantic similarity techniques to calculate the similarity of two contexts
  • Represent context as a graph based data mode and use graph similarity measures to calculate the similarity of two contexts

Related Work

Related Application Areas

Context in Recommender Systems

Mainly there are two ways the context can be incorporated in to recommendation

  • Context Aware Recommendation[5] - extensions of traditional recommendation systems that also take into account contextual condition to recommend items to the user
  • Context Recommendation[9] - recommending to a user the appropriate contexts in which an item should be selected
Important Relevant work
  • Using Context Similarity for Service Recommendation[10]: Usually context aware recommendation considers a single context segment but this will yield to the lack of sufficient recommendations within the context segment. To further improve this, it is important to find out the similarity of the contexts to use similar context rather than restricting to a single context. This approach proposes a method for context similarity when the categories of a context dimensions are defined as concepts in ontology.
Data

Graph Matching for Context Recognition in Ambient Intelligence[6]

Context Information is an important aspect for Ambient Intelligence. This paper proposes an approach to match context pattern of a user against context graphs as a valid method for detecting user's situation and acting up on user's context

Semantic Similarity

Semantic Similarity measures have been used to calculate the similarity of given two entities(can be words or concepts). Most of the existing semantic similarity measures mainly can be categorized into three areas,

  • Text based similarity - Syntactic similarities of words Eg: Levenshtein, Cosine Similarity, Jaccard
  • Corpus based similarity - Leverage information obtained from a corpus Eg: PMI, LSA, ESA, Distributional Similarity
  • Knowledge based similarity - Leverage a semantic network to calculate the similarity, Similarity is based on the shortest path distance of the two entities and/or information content of the least common subsumer

Limitation: Always consider the abstract unit as an entity or word, but this can be a more complex structure than such as multidimensional model or graph based model

Semantic Similarity on RDF Graphs

In general, various graph similarity measures have been proposed in graph theory such as Graph Edit Distance and Subgraph isomorphism, but these techniques do not consider the semantics of the graph stricture. There are couple of existing work which uses graph similarity for the RDF data model. Of them most common areas are,

  • Ontology Alignment

- Graph Matching Approaches for ontology alignments which looks at the structural similarity of the two graphs [1][2]

  • Matching two graph patterns for querying

- Semantic Search via mapping query graph and resource graph [3]

- Path Alignment of the query graph and resource graph [4]

References

[1] Hu, Wei, et al. "Gmo: A graph matching for ontologies." Proceedings of K-CAP Workshop on Integrating Ontologies. 2005.

[2]Tous, Rubén, and Jaime Delgado. "A vector space model for semantic similarity calculation and OWL ontology alignment." Database and Expert Systems Applications. Springer Berlin Heidelberg, 2006.

[3] Jiwei, Haiping Zhu, et al. "An approach for semantic search by matching rdf graphs." In Proceedings of the Special Track on Semantic Web at the 15th International FLAIRS Conference (sponsored by AAAI. 2002.

[4]De Virgilio, Roberto, Antonio Maccioni, and Riccardo Torlone. "A similarity measure for approximate querying over RDF data." Proceedings of the Joint EDBT/ICDT 2013 Workshops. ACM, 2013.

[5]Adomavicius, Gediminas, and Alexander Tuzhilin. "Context-aware recommender systems." Recommender systems handbook. Springer US, 2011. 217-253

[6] Dobrescu, Adrian, and Andrei Olaru. "Graph matching for context recognition." Control Systems and Computer Science (CSCS), 2013 19th International Conference on. IEEE, 2013.

[7] Dey, A. and Abowd, G. (1999) Towards a Better Understanding of Context and Context-Awareness. In Proc. 1st Int. Symp. Handheld and Ubiquitous Computing, Karlsruhe, Germany, pp. 304–307

[8] Dey, A.K. (2001) Understanding and using context. Pers. Ubiquitous Comput., 5, 4–7.

[9] Zheng, Yong, Bamshad Mobasher, and Robin Burke. "Context recommendation using multi-label classification." Web Intelligence (WI) and Intelligent Agent Technologies (IAT), 2014 IEEE/WIC/ACM International Joint Conferences on. Vol. 2. IEEE, 2014.

[10] Liu, Liwei, et al. "Using context similarity for service recommendation." Semantic Computing (ICSC), 2010 IEEE Fourth International Conference on. IEEE, 2010.