Better Caching in Search Advertising Systems with Rapid Refresh Predictions

@article{Li2018BetterCI,
  title={Better Caching in Search Advertising Systems with Rapid Refresh Predictions},
  author={Conglong Li and David G. Andersen and Qiang Fu and Sameh Elnikety and Yuxiong He},
  journal={Proceedings of the 2018 World Wide Web Conference},
  year={2018}
}
To maximize profit and connect users to relevant products and services, search advertising systems use sophisticated machine learning algorithms to estimate the revenue expectations of thousands of matching ad listings per query. These machine learning computations constitute a substantial part of the operating cost, e.g., 10% to 30% of the total gross revenues. It is desirable to cache and reuse previous computation results to reduce this cost, but caching introduces approximation which comes… 
Improving Search Engine Performance Through Dynamic Caching
TLDR
This paper presents a dynamic cache that relies on an online algorithm, which performs a semantic match between the user’s query and queries stored in the cache, and employs a priority queue, where popular queries are stored along with their relevant documents.
Exploiting the Use of Similar Past Search Results Through a Dynamic Cache
TLDR
A dynamic cache based on an on-line algorithm, which takes advantage of similar past queries to reduce the engine’s response times, while simultaneously improving precision, is described.
Improving Approximate Nearest Neighbor Search through Learned Adaptive Early Termination
TLDR
This work builds and train gradient boosting decision tree models to learn and predict when to stop searching for a certain query and applies the learned adaptive early termination to state-of-the-art ANN approaches, and evaluates the end-to-end performance on three million to billion-scale datasets.
Accelerating Deep Learning Classification with Error-controlled Approximate-key Caching
TLDR
A novel caching paradigm is proposed, that is named approximate-key caching, which returns approximate results for lookups of selected input based on cached DL inference results, and an error-correction principled algorithm is named auto-refresh.
Maximizing Page-Level Cache Hit Ratios in LargeWeb Services
TLDR
This work analyzes page requests served by a Microsoft production system and finds that in practice there is potential for improving the page-level hit ratio over existing caching strategies, but that analytically maximizing thepage- level hit ratio is NP-hard.
Kangaroo: Theory and Practice of Caching Billions of Tiny Objects on Flash
TLDR
Kangaroo is a new flash-cache design that optimizes both DRAM usage and flash writes to maximize cache performance while minimizing cost, and is Pareto-optimal across a range of allowed write rates, DRAM sizes, and flash sizes.
Kangaroo: Caching Billions of Tiny Objects on Flash
TLDR
Kangaroo is a new flash-cache design that optimizes both DRAM usage and flash writes to maximize cache performance while minimizing cost, and is Pareto-optimal across a range of allowed write rates, DRAM sizes, and flash sizes.

References

SHOWING 1-10 OF 25 REFERENCES
Workload analysis and caching strategies for search advertising systems
TLDR
This paper presents three domain-specific caching mechanisms that successfully optimize for both revenue and cost in the Bing advertising system and shows that a traditional cache can reduce cost but has negative revenue impact as bad as -14.1%.
Improved techniques for result caching in web search engines
TLDR
A new set of feature-based cache eviction policies that achieve significant improvements over all previous methods, substantially narrowing the existing performance gap to the theoretically optimal (clairvoyant) method.
A refreshing perspective of search engine caching
TLDR
This work introduces a novel algorithm that uses a time-to-live value to set cache entries to expire and selectively refreshes cached results by issuing refresh queries to back-end search clusters, and prioritizes the entries to refresh according to a heuristic that combines the frequency of access with the age of an entry in the cache.
Learning to predict response times for online query scheduling
TLDR
This work investigates the impact of dynamic pruning strategies on query response times, and proposes a framework for predicting the efficiency of a query, and shows that combining multiple efficiency predictors with regression can accurately predict the response time of a queries before it is executed.
Second Chance: A Hybrid Approach for Dynamic Result Caching and Prefetching in Search Engines
TLDR
This work proposes a hybrid result caching strategy where a dynamic result cache is split into two sections: an HTML cache and a docID cache, and proposes a machine learning approach to predict singleton queries, which occur only once in the query stream.
Design trade-offs for search engine caching
TLDR
Using a query log spanning a whole year, it is demonstrated that caching posting lists can achieve higher hit rates than caching query answers and a new algorithm is proposed for static caching of posting lists, which outperforms previous methods.
Caching for Realtime Search
TLDR
It is shown that a classical cache replacement policy, LRU, completely fails to guarantee freshness over time, whereas the proposed CIPs serve 97% of the queries with fresh results, and it is demonstrated that the computational overhead of the algorithms is minor, and that they even allow reducing the cache's memory footprint.
Practical Lessons from Predicting Clicks on Ads at Facebook
TLDR
This paper introduces a model which combines decision trees with logistic regression, outperforming either of these methods on its own by over 3%, an improvement with significant impact to the overall system performance.
Ad click prediction: a view from the trenches
TLDR
The goal of this paper is to highlight the close relationship between theoretical advances and practical engineering in this industrial setting, and to show the depth of challenges that appear when applying traditional machine learning methods in a complex dynamic system.
Web-Scale Bayesian Click-Through rate Prediction for Sponsored Search Advertising in Microsoft's Bing Search Engine
TLDR
A new Bayesian click-through rate (CTR) prediction algorithm used for Sponsored Search in Microsoft's Bing search engine is described, based on a probit regression model that maps discrete or real-valued input features to probabilities.
...
...