Connecting Large Language Models (LLMs) with the ability to leverage APIs (Web Search, Charting, Calculators, Calendar, Flight Search, Hotel Search, Data Lookup, etc. ) is likely to allow us to solve a variety of new hard problems.Several research efforts have made this observation and suggested recipes for LLMs to emit API calls, and proposed mechanisms by which they can generate additional text conditioned on the output for the API call.However, in practice, the focus has been on relatively simple slot-filling tasks that make an API call rather unlocking novel capabilities by combining different tools, reasoning over the response from a tool, making multiple invocations, or complex planning.In this paper, we pose the following question: what does it mean to say that an LLM is proficient at using a set of APIs?We answer this question in the context of structured APIs by defining seven capabilities for API-use.We provide an approach for generating synthetic tasks that exercise each of these capabilities given only the description of an API 1 .We argue that this provides practitioners with a principled way to construct a dataset to evaluate an LLM's ability to use a given set of APIs.Through human evaluations, we show that our approach produces high-quality tasks for each of the seven capabilities.We also describe how we used this approach to on-board new API and create principled evaluation sets for multiple LLM-based products.

Attribute value extraction refers to the task of identifying values of an attribute of interest from product information. Product attribute values are essential in many e-commerce scenarios, such as customer service robots, product ranking, retrieval and recommendations. While in the real world, the attribute values of a product are usually incomplete and vary over time, which greatly hinders the practical applications. In this paper, we introduce MAVE, a new dataset to better facilitate research on product attribute value extraction. MAVE is composed of a curated set of 2.2 million products from Amazon pages, with 3 million attribute-value annotations across 1257 unique categories. MAVE has four main and unique advantages: First, MAVE is the largest product attribute value extraction dataset by the number of attribute-value examples. Second, MAVE includes multi-source representations from the product, which captures the full product information with high attribute coverage. Third, MAVE represents a more diverse set of attributes and values relative to what previous datasets cover. Lastly, MAVE provides a very challenging zero-shot test set, as we empirically illustrate in the experiments. We further propose a novel approach that effectively extracts the attribute value from the multi-source product information. We conduct extensive experiments with several baselines and show that MAVE is an effective dataset for attribute value extraction task. It is also a very challenging task on zero-shot attribute extraction. Data is available at \urlhttps://github.com/google-research-datasets/MAVE .

This paper explores learning rich self-supervised entity representations from large amounts of associated text. Once pre-trained, these models become applicable to multiple entity-centric tasks such as ranked retrieval, knowledge base completion, question answering, and more. Unlike other methods that harvest self-supervision signals based merely on a local context within a sentence, we radically expand the notion of context to include any available text related to an entity. This enables a new class of powerful, high-capacity representations that can ultimately distill much of the useful information about an entity from multiple text sources, without any human supervision. We present several training strategies that, unlike prior approaches, learn to jointly predict words and entities – strategies we compare experimentally on downstream tasks in the TV-Movies domain, such as MovieLens tag prediction from user reviews and natural language movie search. As evidenced by results, our models match or outperform competitive baselines, sometimes with little or no fine-tuning, and are also able to scale to very large corpora. Finally, we make our datasets and pre-trained models publicly available. This includes Reviews2Movielens, mapping the ~1B word corpus of Amazon movie reviews (He and McAuley, 2016) to MovieLens tags (Harper and Konstan, 2016), as well as Reddit Movie Suggestions with natural language queries and corresponding community recommendations.

Attribute value extraction refers to the task of identifying values of an attribute of interest from product information. It is an important research topic which has been widely studied in e-Commerce and relation learning. There are two main limitations in existing attribute value extraction methods: scalability and generalizability. Most existing methods treat each attribute independently and build separate models for each of them, which are not suitable for large scale attribute systems in real-world applications. Moreover, very limited research has focused on generalizing extraction to new attributes.

In this paper, we consider the problem of constructing a comprehensive database of events taking place around the world. Events include small hyper-local events like farmer's markets, neighborhood garage sales, as well as larger concerts and festivals. Designing a high-precision and high-recall event extractor from unstructured pages across the whole web is a challenging problem. We cannot resort overly to domain-specific strategies since it needs to work on all web pages, including on new domains; we need to account for variations in page layouts and structure across websites. Further, we need to deal with low-quality pages on the web with limited structure. We have built an ML-powered extraction system to solve this problem, using schema.org annotations as training data. Our extraction system operates in two phases. In the first phase, we generate raw event information from individual web pages. To do this, an \em event page classifier predicts if a web page contains any event information; this is then followed by a \em single/multiple classifier that decides if the page contains a single event or multiple events; the first phase concludes by applying \em event extractors that extract the key fields of a public event (the title, the date/time information, and the location information). In the second phase, we further improve the extraction quality via three novel algorithms, \em repeated patterns, \em event consolidation and \em wrapper induction, which are designed to use the raw event extractions as input and generate events whose quality is significantly higher. We evaluate our extraction models on two large scale publicly available web corpus, Common Crawl and ClueWeb12. Experimental analysis shows that our methodology achieves over 95% extraction precision and recall on both datasets.

Recommender systems are a key technology for many online services including e-commerce, movies, music, and news. Online retailers use product recommender systems to help users discover items that they may like. However, building a large-scale product recommender system is a challenging task. The problems of sparsity and cold-start are much more pronounced in this domain. Large online retailers have used good recommendations to drive user engagement and improve revenue, but the complexity involved is a roadblock to widespread adoption by smaller retailers. In this paper, we tackle the problem of generating product recommendations for tens of thousands of online retailers. Sigmund is an industrial-scale system for providing recommendations as a service. Sigmund was deployed to production in early 2014 and has been serving retailers every day. We describe the design choices that we made in order to train accurate matrix factorization models at minimal cost. We also share the lessons we learned from this experience - both from a machine learning perspective and a systems perspective. We hope that these lessons are useful for building future machine-learning services.

In recent years, interest in recommender research has shifted from explicit feedback towards implicit feedback data. A diversity of complex models has been proposed for a wide variety of applications. Despite this, learning from implicit feedback is still computationally challenging. So far, most work relies on stochastic gradient descent (SGD) solvers which are easy to derive, but in practice challenging to apply, especially for tasks with many items. For the simple matrix factorization model, an efficient coordinate descent (CD) solver has been previously proposed. However, efficient CD approaches have not been derived for more complex models.

Audience selection is a key problem in display advertising systems in which we need to select a list of users who are interested (i.e., most likely to buy) in an advertising campaign. The users' past feedback on this campaign can be leveraged to construct such a list using collaborative filtering techniques such as matrix factorization. However, the user-campaign interaction is typically extremely sparse, hence the conventional matrix factorization does not perform well. Moreover, simply combining the users feedback from all campaigns does not address this since it dilutes the focus on target campaign in consideration. To resolve these issues, we propose a novel focused matrix factorization model (FMF) which learns users' preferences towards the specific campaign products, while also exploiting the information about related products. We exploit the product taxonomy to discover related campaigns, and design models to discriminate between the users' interest towards campaign products and non-campaign products. We develop a parallel multi-core implementation of the FMF model and evaluate its performance over a real-world advertising dataset spanning more than a million products. Our experiments demonstrate the benefits of using our models over existing approaches.

Items in recommender systems are usually associated with annotated attributes: for e.g., brand and price for products; agency for news articles, etc. Such attributes are highly informative and must be exploited for accurate recommendation. While learning a user preference model over these attributes can result in an interpretable recommender system and can hands the cold start problem, it suffers from two major drawbacks: data sparsity and the inability to model random effects. On the other hand, latent-factor collaborative filtering models have shown great promise in recommender systems; however, its performance on rare items is poor. In this paper we propose a novel model LFUM, which provides the advantages of both of the above models. We learn user preferences (over the attributes) using a personalized Bayesian hierarchical model that uses a combination(additive model) of a globally learned preference model along with user-specific preferences. To combat data-sparsity, we smooth these preferences over the item-taxonomy using an efficient forward-filtering and backward-smoothing inference algorithm. Our inference algorithms can handle both discrete attributes (e.g., item brands) and continuous attributes (e.g., item prices). We combine the user preferences with the latent-factor models and train the resulting collaborative filtering system end-to-end using the successful BPR ranking algorithm. In our extensive experimental analysis, we show that our proposed model outperforms several commonly used baselines and we carry out an ablation study showing the benefits of each component of our model.

Recommender systems based on latent factor models have been effectively used for understanding user interests and predicting future actions. Such models work by projecting the users and items into a smaller dimensional space, thereby clustering similar users and items together and subsequently compute similarity between unknown user-item pairs. When user-item interactions are sparse ( sparsity problem) or when new items continuously appear ( cold start problem), these models perform poorly. In this paper, we exploit the combination of taxonomies and latent factor models to mitigate these issues and improve recommendation accuracy. We observe that taxonomies provide structure similar to that of a latent factor model: namely, it imposes human-labeled categories (clusters) over items. This leads to our proposed taxonomy-aware latent factor model (TF) which combines taxonomies and latent factors using additive models. We develop efficient algorithms to train the TF models, which scales to large number of users/items and develop scalable inference/recommendation algorithms by exploiting the structure of the taxonomy. In addition, we extend the TF model to account for the temporal dynamics of user interests using high-order Markov chains . To deal with large-scale data, we develop a parallel multi-core implementation of our TF model. We empirically evaluate the TF model for the task of predicting user purchases using a real-world shopping dataset spanning more than a million users and products. Our experiments demonstrate the benefits of using our TF models over existing approaches, in terms of both prediction accuracy and running time.

Probabilistic database systems have successfully established themselves as a tool for managing uncertain data. However, much of the research in this area has focused on efficient query evaluation and has largely ignored two key issues that commonly arise in uncertain data management: First, how to provide explanations for query results, e.g., Why is this tuple in my result? or Why does this output tuple have such high probability?. Second, the problem of determining the sensitive input tuples for the given query, e.g., users are interested to know the input tuples that can substantially alter the output, when their probabilities are modified (since they may be unsure about the input probability values). Existing systems provide the lineage/provenance of each of the output tuples in addition to the output probabilities, which is a boolean formula indicating the dependence of the output tuple on the input tuples. However, lineage does not immediately provide a quantitative relationship and it is not informative when we have multiple output tuples. In this paper, we propose a unified framework that can handle both the issues mentioned above to facilitate robust query processing. We formally define the notions of influence and explanations and provide algorithms to determine the top-l influential set of variables and the top-l set of explanations for a variety of queries, including conjunctive queries, probabilistic threshold queries, top-k queries and aggregation queries. Further, our framework naturally enables highly efficient incremental evaluation when input probabilities are modified (e.g., if uncertainty is resolved). Our preliminary experimental results demonstrate the benefits of our framework for performing robust query processing over probabilistic databases.

In this paper, we address the problem of scalably evaluating conjunctive queries over correlated probabilistic databases containing tuple or attribute uncertainties. Like previous work, we adopt a two-phase approach where we first compute lineages of the output tuples, and then compute the probabilities of the lineage formulas. However unlike previous work, we allow for arbitrary and complex correlations to be present in the data, captured via a forest of junction trees. We observe that evaluating even read-once (tree structured) lineages (e.g., those generated by hierarchical conjunctive queries), polynomially computable over tuple independent probabilistic databases, is #P-complete for lightly correlated probabilistic databases like Markov sequences. We characterize the complexity of exact computation of the probability of the lineage formula on a correlated database using a parameter called lwidth (analogous to the notion of treewidth). For lineages that result in low lwidth, we compute exact probabilities using a novel message passing algorithm, and for lineages that induce large lwidths, we develop approximate Monte Carlo algorithms to estimate the result probabilities. We scale our algorithms to very large correlated probabilistic databases using the previously proposed INDSEP data structure. To mitigate the complexity of lineage evaluation, we develop optimization techniques to process a batch of lineages by sharing computation across formulas, and to exploit any independence relationships that may exist in the data. Our experimental study illustrates the benefits of using our algorithms for processing lineage formulas over correlated probabilistic databases.

In this paper, we address the problem of efficient query evaluation over highly correlated probabilistic streams. We observe that although probabilistic streams tend to be strongly correlated in space and time, the correlations are usually quite structured (i.e., the same set of dependencies and independences repeat across time) and Markovian (i.e., the state at time "t+1" is independent of the states at previous times given the state at time "t"). We exploit this observation to compactly encode probabilistic streams by decoupling the correlation structure (the set of dependencies) from the actual probability values. We develop novel stream processing operators that can efficiently and incrementally process new data items; our operators are based on the previously proposed framework of viewing probabilistic query evaluation as inference over probabilistic graphical models (PGMs) [P. Sen and A. Deshpande, 2007]. We develop a query planning algorithm that constructs efficient query plans that are executable in polynomial-time whenever possible, and we characterize queries for which such plans are not possible. Finally we conduct an extensive experimental evaluation that illustrates the advantages of exploiting the structured nature of correlations in probabilistic streams.

With large amounts of correlated probabilistic data being generated in a wide range of application domains including sensor networks, information extraction, event detection etc., effectively managing and querying them has become an important research direction. While there is an exhaustive body of literature on querying independent probabilistic data, supporting efficient queries over large-scale, correlated databases remains a challenge. In this paper, we develop efficient data structures and indexes for supporting inference and decision support queries over such databases. Our proposed hierarchical data structure is suitable both for in-memory and disk-resident databases. We represent the correlations in the probabilistic database using a junction tree over the tuple-existence or attribute-value random variables, and use tree partitioning techniques to build an index structure over it. We show how to efficiently answer inference and aggregation queries using such an index, resulting in orders of magnitude performance benefits in most cases. In addition, we develop novel algorithms for efficiently keeping the index structure up-to-date as changes (inserts, updates) are made to the probabilistic database. We present a comprehensive experimental study illustrating the benefits of our approach to query processing in probabilistic databases.

In this paper, we address the problem of extending a relational database system to facilitate efficient real-time application of dynamic probabilistic models to streaming data. We use the recently proposed abstraction of model-based views for this purpose, by allowing users to declaratively specify the model to be applied, and by presenting the output of the models to the user as a probabilistic database view. We support declarative querying over such views using an extended version of SQL that allows for querying probabilistic data. Underneath we use particle filters, a class of sequential Monte Carlo algorithms, to represent the present and historical states of the model as sets of weighted samples (particles) that are kept up-to-date as new data arrives. We develop novel techniques to convert the queries on the model-based view directly into queries over particle tables, enabling highly efficient query processing. Finally, we present experimental evaluation of our prototype implementation over several synthetic and real datasets, that demonstrates the feasibility of online modeling of streaming data using our system and establishes the advantages of tight integration between dynamic probabilistic models and databases.