Search efficiency and serving efficiency are two major axes in building feature interactions and expediting the model development process in recommender systems. Searching for the optimal feature interaction design on large-scale benchmarks requires extensive cost due to the sequential workflow on the large volume of data. In addition, fusing interactions of various sources, orders, and mathematical operations introduces potential conflicts and additional redundancy toward recommender models, leading to sub-optimal trade-offs in performance and serving cost. This paper presents DistDNAS as a neat solution to brew swift and efficient feature interaction design. DistDNAS proposes a supernet incorporating interaction modules of varying orders and types as a search space. To optimize search efficiency, DistDNAS distributes the search and aggregates the choice of optimal interaction modules on varying data dates, achieving a speed-up of over 25× and reducing the search cost from 2 days to 2 hours. To optimize serving efficiency, DistDNAS introduces a differentiable cost-aware loss to penalize the selection of redundant interaction modules, enhancing the efficiency of discovered feature interactions in serving. We extensively evaluate the best models crafted by DistDNAS on a 1TB Criteo Terabyte dataset. Experimental evaluations demonstrate 0.001 AUC improvement and 60% FLOPs saving over current state-of-the-art CTR models.

Motif discovery in sequence data is fundamental to many biological problems such as antibody biomarker identification. Recent advances in instrumental techniques make it possible to generate thousands of protein sequences at once, which raises a big data issue for the existing motif finding algorithms: They either work only in a small scale of several hundred sequences or have to trade accuracy for efficiency. In this work, we demonstrate that by intelligently clustering sequences, it is possible to significantly improve the scalability of all the existing motif finding algorithms without losing accuracy at all. An anchor based sequence clustering algorithm (ASC) is thus proposed to divide a sequence dataset into multiple smaller clusters so that sequences sharing the same motif will be located into the same cluster. Then an existing motif finding algorithm can be applied to each individual cluster to generate motifs. In the end, the results from multiple clusters are merged together as final output. Experimental results show that our approach is generic and orders of magnitude faster than traditional motif finding algorithms. It can discover motifs from protein sequences in the scale that no existing algorithm can handle. In particular, ASC reduces the running time of a very popular motif finding algorithm, MEME, from weeks to a few minutes with even better accuracy.

Given a graph query Q posed on a knowledge graph G, top-k graph querying is to find k matches in G with the highest ranking score according to a ranking function. Fast top-k search in knowledge graphs is challenging as both graph traversal and similarity search are expensive. Conventional top-k graph search is typically based on threshold algorithm (TA), which can no long fit the demand in the new setting. This work proposes STAR, a top-k knowledge graph search framework. It has two components: (a) a fast top-k algorithm for star queries, and (b) an assembling algorithm for general graph queries. The assembling algorithm uses star query as a building block and iteratively sweeps the star match lists with a dynamically adjusted bound. For top-k star graph query where an edge can be matched to a path with bounded length d, we develop a message passing algorithm, achieving time complexity O(d2|E| + md) and space complexity linear to d|V| (assuming the size of Q and k is bounded by a constant), where m is the maximum node degree in G. STAR can further be leveraged to answer general graph queries by decomposing a query to multiple star queries and joining their results later. Learning-based techniques to optimize query decomposition are also developed. We experimentally verify that STAR is 5-10 times faster than the state-of-the-art TA-style graph search algorithm, and 10-100 times faster than a belief propagation approach.

Collaborative networks are common in real life, where domain experts work together to solve tasks issued by customers. How to model the proficiency of experts is critical for us to understand and optimize collaborative networks. Traditional expertise models, such as topic model based methods, cannot capture two aspects of human expertise simultaneously: Specialization (what area an expert is good at?) and Proficiency Level (to what degree?). In this paper, we propose new models to overcome this problem. We embed all historical task data in a lower dimension space and learn vector representations of expertise based on both solved and unsolved tasks. Specifically, in our first model, we assume that each expert will only handle tasks whose difficulty level just matches his/her proficiency level, while experts in the second model accept tasks whose levels are equal to or lower than his/her proficiency level. Experiments on real world datasets show that both models outperform topic model based approaches and standard classifiers such as logistic regression and support vector machine in terms of prediction accuracy. The learnt vector representations can be used to compare expertise in a large organization and optimize expert allocation.

As a fundamental technique for graph analysis, graph kernels have been successfully applied to a wide range of problems. Unfortunately, the high computational complexity of existing graph kernels is limiting their further applications to larger-scale graph datasets. In this paper, we propose a fast graph kernel, the descriptor matching (DM) kernel, for graphs with both categorical and numerical attributes. The computation time of the DM kernel is linear with respect to graph size. On graphs with n nodes and m edges, the kernel computation for two graphs can be done in O(n+m) time. Although there are other linear-time graph kernels, most of them are restricted to graphs with only categorical attributes; their efficiency mainly comes from the sparseness of the feature space resulted from the mutually orthogonal categorical attributes. Extensive experiments on both synthetic and real-world graph datasets show promising performance of DM in both accuracy and efficiency: On graphs with both categorical and numerical attributes, DM is orders of magnitude faster than several state-of-the-art graph kernels, while being much more accurate than the only graph kernel that is more efficient.

Named entity disambiguation is the task of disambiguating named entity mentions in natural language text and link them to their corresponding entries in a knowledge base such as Wikipedia. Such disambiguation can help enhance readability and add semantics to plain text. It is also a central step in constructing high-quality information network or knowledge graph from unstructured text. Previous research has tackled this problem by making use of various textual and structural features from a knowledge base. Most of the proposed algorithms assume that a knowledge base can provide enough explicit and useful information to help disambiguate a mention to the right entity. However, the existing knowledge bases are rarely complete (likely will never be), thus leading to poor performance on short queries with not well-known contexts. In such cases, we need to collect additional evidences scattered in internal and external corpus to augment the knowledge bases and enhance their disambiguation power. In this work, we propose a generative model and an incremental algorithm to automatically mine useful evidences across documents. With a specific modeling of "background topic" and "unknown entities", our model is able to harvest useful evidences out of noisy information. Experimental results show that our proposed method outperforms the state-of-the-art approaches significantly: boosting the disambiguation accuracy from 43% (baseline) to 86% on short queries derived from tweets.

Massively parallel DNA sequencing technologies are revolutionizing genomics research. Billions of short reads generated at low costs can be assembled for reconstructing the whole genomes. Unfortunately, the large memory footprint of the existing de novo assembly algorithms makes it challenging to get the assembly done for higher eukaryotes like mammals. In this work, we investigate the memory issue of constructing de Bruijn graph, a core task in leading assembly algorithms, which often consumes several hundreds of gigabytes memory for large genomes. We propose a disk-based partition method, called Minimum Substring Partitioning (MSP), to complete the task using less than 10 gigabytes memory, without runtime slowdown. MSP breaks the short reads into multiple small disjoint partitions so that each partition can be loaded into memory, processed individually and later merged with others to form a de Bruijn graph. By leveraging the overlaps among the k-mers (substring of length k), MSP achieves astonishing compression ratio: The total size of partitions is reduced from Θ(kn) to Θ(n), wherenis the size of the short read database, andkis the length of ak-mer. Experimental results show that our method can build de Bruijn graphs using a commodity computer for any large-volume sequence dataset.