Graduate School and Research Center in Digital Sciences

Impact of popular content relational structure on joint caching and recommendation policies

Costantini, Marina; Spyropoulos, Thrasyvoulos

WIOPT 2020, Workshop on Content Caching and Delivery in Wireless Networks (CCDWN), at the 18th International Symposium on Modeling and Optimization in Mobile, Ad Hoc and Wireless Networks, 15-19 June 2020, Volos, Greece (Virtual Conference)

Recent work has shown that the performance of caching systems can be boosted by the delivery of alternative but related content. In this setup, recommendation systems are exploited to offer the user appealing alternatives if the original request is not found in the cache. This framework relies on the assumption that a content can partially or completely substitute another if they are sufficiently similar. In this work we model these similarity relations as a graph, where each content is a node and related contents are linked by an edge. We then study how the characteristics of this content graph constrain the gains of designing jointly the caching and the recommendations with respect to just using a simple baseline in the soft cache hits setup. We start by selecting a number of descriptive graph features that we expect to play a role in the performance of related content delivery policies. We then analyze the effect of each of these features on the policies’ performance through extensive simulations with synthetic data. Our results confirm that features such as the degree distribution and clustering coefficient of the graph are crucial to decide whether the close-to-optimal algorithm will perform significantly better than the simple baseline. Our experiments with four real-world distinct datasets further support these observations. Motivated by these clear dependencies, we conclude by showing that we can train a classifier to predict the gains attainable by a “smart” policy with respect to a low-complexity baseline using only a few m acroscopic graph features as predictor variables. 

Document Bibtex

Title:Impact of popular content relational structure on joint caching and recommendation policies
Type:Conference
Language:English
City:Volos
Country:GREECE
Date:
Department:Communication systems
Eurecom ref:6272
Copyright: © IFIP. Personal use of this material is permitted. The definitive version of this paper was published in WIOPT 2020, Workshop on Content Caching and Delivery in Wireless Networks (CCDWN), at the 18th International Symposium on Modeling and Optimization in Mobile, Ad Hoc and Wireless Networks, 15-19 June 2020, Volos, Greece (Virtual Conference) and is available at :
Bibtex: @inproceedings{EURECOM+6272, year = {2020}, title = {{I}mpact of popular content relational structure on joint caching and recommendation policies}, author = {{C}ostantini, {M}arina and {S}pyropoulos, {T}hrasyvoulos}, booktitle = {{WIOPT} 2020, {W}orkshop on {C}ontent {C}aching and {D}elivery in {W}ireless {N}etworks ({CCDWN}), at the 18th {I}nternational {S}ymposium on {M}odeling and {O}ptimization in {M}obile, {A}d {H}oc and {W}ireless {N}etworks, 15-19 {J}une 2020, {V}olos, {G}reece ({V}irtual {C}onference)}, address = {{V}olos, {GREECE}}, month = {06}, url = {http://www.eurecom.fr/publication/6272} }
See also: