Google Inc.

Jayant Madhavan

Alon Yitzchak Halevy

Sreeram Viswanath Balakrishnan

John Blitzer

Ibrahim Badr

Srinivasan Venkatachary

Matthew K Gray

David H Ko

Hongrae Lee

Mahsan Rofouei

Warren H Y Shen

Sam Oates

Aditya Shah

Phiroze Parakh

Corinn Herrick

Anno Langen

Hector Gonzalez

Eldar Sadikov

Fei Wu

Hazem Elmeleegy

Andrin Von Richenberg

Nitin Kumar Gupta

Preyas Popat

Andrin Von Rechenberg

Aliaksei Severyn

Gaurav Bhaskar Gite

Tuan Anh Cao

Techniques of generating short answers for queries by a search engine include performing a training operation on a corpus of training data to train the score prediction engine, the corpus of training data including candidate passages providing short answers for display in callouts and remaining respective passages, from which a top scoring short answer is generated. In such implementations, the corpus of training data further includes the remaining respective passages and the respective titles of the candidate passage and remaining respective passages.

Multi source extraction and scoring of short query answers

At least selectively utilizing a large language model (LLM) in generating a natural language (NL) based summary to be rendered in response to a query. In some implementations, in generating the NL based summary additional content is processed using the LLM. The additional content is in addition to query content of the query itself and, in generating the NL based summary, can be processed using the LLM and along with the query content—or even independent of the query content. Processing the additional content can, for example, mitigate occurrences of the NL based summary including inaccuracies and/or can mitigate occurrences of the NL based summary being over-specified and/or under-specified.

Generative summaries for search results

In one aspect, a method includes receiving a query determined to be a question query that seeks an answer response and data identifying resources determined to be responsive to the query; identifying structured content set in a top-ranked subset of the resources, each structured content set being content arranged according to related attributes in one of the resources; for each identified structured content set, determining whether the query matches the structured content set based on terms of the query matching related attributes of the structured content set; selecting one of the structured content sets for which the query is determined to match; generating, from the selected structured content set, a structured fact set from the related attributes that matched the terms of the query; and providing the structured fact set with search results that identify the resources determined to be responsive to the query.

Answer facts from structured content

At least selectively utilizing a large language model (LLM) in generating a natural language (NL) based summary to be rendered in response to a query. In some implementations, in generating the NL based summary additional content is processed using the LLM. The additional content is in addition to query content of the query itself and, in generating the NL based summary, can be processed using the LLM and along with the query content&#x2014;or even independent of the query content. Processing the additional content can, for example, mitigate occurrences of the NL based summary including inaccuracies and/or can mitigate occurrences of the NL based summary being over-specified and/or under-specified.

Methods, systems, and apparatus, including computer programs encoded on computer storage media, for clustering query refinements. One method includes building a representation of a graph for a first query, wherein the graph has a node for the first query, a node for each of a plurality of refinements for the first query, and a node for each document in the document sets of the refinements, and wherein the graph has edges from the first query node to each of the refinement nodes, edges from the first query to each document in the respective document set of the first query, edges from each refinement to each document in the respective document set of the refinement, and edges from each refinement to each co-occurring query of the refinement. The method further includes clustering the refinements into refinement clusters by partitioning the refinement nodes in the graph into proper subsets.

Clustering query refinements by inferred user intent

Methods, systems, and apparatus, including computer program products, for generating aspects associated with entities. In some implementations, a method includes receiving data identifying an entity; generating a group of candidate aspects for the entity; modifying the group of candidate aspects to generate a group of modified candidate aspects comprising combining similar candidate aspects and grouping candidate aspects using one or more aspect classes each associated with one or more candidate aspects; ranking one or more modified candidate aspects in the group of modified candidate aspects based on a diversity score and a popularity score; and storing an association between one or more highest ranked modified candidate aspects and the entity. The aspects can be used to organize and present search results in response to queries for the entity.

Identifying query aspects

Systems and techniques are provided for receiving an input column and a search keyword and providing one or more suggested columns with which to merge the input column. A coverage score and a refinity score are calculated for potential columns based on the input column as well as a search score based on the search keyword. The one or more suggested columns may be determined based on the coverage score, refinity score, and/or the search score. The input column and/or a potential column may be modified based on a function and the modification may result in a plurality of modified input and/or potential columns. Coverage, refinity, and search scores may be calculated based on the modified columns.

Searching for join candidates

Aspects of the invention provide a service for data management and integration across a wide range of applications. Clustered computers may be arranged in a cloud-type configuration for storing and handling large amounts of user data under the control of a front-end management server. Communities of distributed users may collaborate on the data across multiple enterprises. Very large tabular data files are uploaded to the storage facilities. The data files are maintained as tables, and a composite table of related information is created and maintained in response to user queries. Different ways of visualizing the data are provided. Depending on the amount of information that can be displayed, features in a spatial index may the thinned for presentation. Spatial and structured queries are processing and results are intersected to obtain information for display.

Scalable rendering of large spatial databases

List information can be extracted into database tables. A number of fields are independently determined for items in list. A number of database table columns are determined from most common number of list item fields. New fields are determined for items with more fields than database columns. Null fields are inserted into items with fewer fields than database columns. Information from items having the same number of fields as database columns is written to database table rows. Information from each field is written to a corresponding database table column. Streaks of poorly matching cells in a database table row are determined. Streak cells are merged and new cells are determined. Null cells are inserted if number of new cells is less than number of cells in the streak. Information from the new cells is written to the table row and columns that define the streak.

Harvesting relational tables from lists on the web

Growing community of inventors

San Francisco, CA, United States of America

Jayant Madhavan