Baidu Online Network Technology (beijing) Co., Ltd.

Gaolin Fang

Hongjian Zou

Jun Cheng

Libo Fu

Yu Chen

Wei Xu

Heng Luo

Jia Gui

Xiaoguang Gu

Hailang Liu

Embodiments of the present disclosure disclose a method, apparatus, server and storage medium for image retrieval. The method includes: identifying a plurality of groups of images having identical contents from images on all webpages; aggregating, for each image group, image-related texts on all source webpages of each image to obtain text descriptions of each image group; establishing an inverted index for each image in the image groups based on the text descriptions of the image group, the inverted index at least including, for each text description, source webpages corresponding to all text descriptions of the image group of the text description; and performing image retrieval based on an inputted query and the inverted index.

Method, apparatus, server and storage medium for image retrieval

The present invention provides a model generation method, word weighting method, device, apparatus, and computer storage medium. In one aspect, the embodiments of the present invention obtain other images that are identical to a specified image, and use the specified image and said other images as sample images; and then obtain a text clustering based on the texts of the sample images; and then obtain text features based on the text clustering and obtaining visual features based on the sample images; and carry out machine learning based on the text features and the visual features to generate a first regression model and a first sorting model for word weighting on words in the text of an image. The embodiments of the present invention therefore can improve the accuracy of the word weighting result in an image text, thus improving the accuracy of an image search result.

Model generation method, word weighting method, device, apparatus, and computer storage medium

A search method and apparatus. A specific embodiment of the search method includes: finding a set of first time-effective search queries from a search log; selecting search queries meeting one of the following selection requirements as candidate time-effective search queries: included in the search log and semantically associated with the first time-effective search queries in the set of first time-effective search queries; included in the search log and containing a combination of preset keywords; processing the candidate time-effective search queries to obtain second time-effective search queries; performing search with a second time-effective search query, when a search query entered by the user matches the second time-effective search query. The present disclosure achieves obtaining the second time-effective search queries from the search queries of the search log based on the identified first time-effective search queries, thus raising the identification recall rate and further improving the identification effect of the time-effective search queries, on the condition that the identification accuracy is ensured in the identification process of the time-effective search queries.

Search method and apparatus

An extended search method and apparatus is provided. An interest term model of each user is established. An extended term model of each fourth search sequence is established. A corresponding extended search term is determined based on a current search sequence of a current user and based on an interest term model of the current user as well as an extended term model of the current search sequence; and a corresponding search result is provided for the current user based on the current search sequence and the extended search term. The present invention can implement simpler and more efficient personalized searches, and is advantageous in terms of meeting long-tail requirements of users (that is, adding personalized result recalling), reducing calculation and storage costs, universality, practicability, and the like.

Extended search method and apparatus

Embodiments of the present invention disclose a method and apparatus of generating an image characteristic representation of a query, and an image search method and apparatus. The method of generating an image characteristic representation of a query comprises: acquiring a clicked image set corresponding to a target query based on an image click log of a search user; generating image characteristic vectors corresponding to clicked images in the clicked image set based on image content characteristics of the clicked images; and clustering the clicked images based on the image characteristic vectors, and using a clustering result as an image characteristic representation of the target query.

Method and apparatus of generating image characteristic representation of query, and image search method and apparatus

Embodiments of the present invention disclose a query result bottom retrieval method and apparatus. The query result bottom retrieval method includes: acquiring, from a resource library and based on a user input target query, query resources associated with the user input target query; acquiring comparative scoring characteristics of each of the query resources, the comparative scoring characteristics including base relevance characteristics; inputting the comparative scoring characteristics of each of the query resources into a pre-trained rank model, to acquire relevance scoring values corresponding to the query resources and output by the rank model, the rank model being a GBRank model; and ranking the query resources based on the relevance scoring values, and selecting, according to the ranking result, a set number of target resources as a bottom retrieval result corresponding to the target query. The technical solution of the present invention, by using a GBRank rank model, optimizes a traditional query result bottom retrieval method, and enhances relevance between a retrieved target resource and a target query.

Query result bottom retrieval method and apparatus

Embodiments of the present invention disclose a method and an apparatus of establishing an image search relevance prediction model, and an image search method and apparatus. The method of establishing an image search relevance prediction model comprises: training a pre-constructed original deep neural network by using a training sample, wherein the training sample comprises: a query and image data, and the original deep neural network comprises: a representation vector generation network and a relevance calculation network; and using the trained original deep neural network as the image search relevance prediction model. The technical solution of the present invention optimizes the existing image search technology, and has stronger capabilities than the prior art as well as various integrations and variations in terms of semantic matching between a query and an image text, semantic matching between a query and image content, click generalization and the like.

Method and apparatus of establishing image search relevance prediction model, and image search method and apparatus

Search engine method includes: receiving a user query request input; searching candidate results matching with the query request; determining a semantic relatedness between the query request and each candidate result based on a click-escape model; sorting the candidate results according to the semantic relativity. The click-escape model has an escape dictionary, a non-escape dictionary, or a combination thereof. Sorting candidate results of a search in accordance with a semantic relatedness can enhance the sorting effect of the searched results, avoid searched results which do not match the user's query appearing in the forefront of the searched result list, and guarantee a good user experience.

Growing community of inventors

Beijing, China

Gaolin Fang