Xerox Corporation

Naver Corporation

Conduent Business Services, LLC

Jean-Michel Renders

Caroline Privault

Eric Gaussier

Cyril Goutte

Marc Dymetman

Victor Ciriza

Yves Hoppenot

Jacki O'Neill

Matthias Galle

Amin Mantrach

Eric H Cheminot

Ana Fucs

Sriram Venkatapathy

Matthias Gallé

Francois Pacull

Damien Cramet

Julien Ah-Pine

Guillaume Jacquet

Ye Deng

Grégoire Gerard

Sruthi Viswanathan

Irina Matveeva

François Ragnet

Bruno Cavestro

Viet Ha-Thuc

Mathieu Knibiehly

Gregory Bauduin

Marie-Luce Viaud

Spandana Gella

Ludovic Menuge

Phong Le

Thibault Formal

Bernard Omidvar-Tehrani

Thibaut Thonet

Till Kletti

Yagmur Gizem Cinar

A method of ranking items for a given entity uses sets of triplets <u, i, j>, each set of triplets including an entity u and a pair of items i and j with a known relative relevance for entity u, to train a learnable scoring function ƒ and to learn optimized values of a first set θ of learnable parameters. The training includes optimizing a loss function depending on θ, on a second set of learnable parameters θ, and on a probability of having the item i preferred to the item j by the entity u. The probability defines a continuum between pointwise and pairwise ranking of items through a learnable mixing function depending on θ. After training, the trained learnable scoring function ƒ is applied to all input pairs <u', i′> to rank all items i′ for an entity u′.

Adaptive pointwise-pairwise learning to rank

A method of training a hypergraph convolutional network (HGCN) includes: receiving training data including search instances and recommendation instances; constructing a hypergraph from the training data, where each node of the hypergraph represents one of a user profile, a query term, and a content item, and where the hypergraph represents each of the search instances and each of the recommendation instances as a hyperedge linking corresponding ones of the nodes; initializing base embeddings associated with the hypergraph nodes; propagating the base embeddings through one or more convolutional layers of the HGCN to obtain, for each of the convolutional layers, respective embeddings of the nodes of the hypergraph; computing, based on the base embeddings and the respective embeddings obtained from each of the one or more convolutional layers: a first loss; and a second loss; and selectively updating ones of the base embeddings based on the first and second losses.

Joint personalized search and recommendation with hypergraph convolutional networks

A method for ranking a set of objects includes: receiving the set of objects to rank, a relevance score for each object, and a set of objective functions; based on the relevance scores for the objects, defining a decision space having n decision variables using a polytope, where n is the number of objects to rank and where vertices of the polytope represent permutations of exposures provided to the objects in the set by corresponding rankings; determining a Pareto-set for the set of objective functions; based on a Pareto-optimal point in the Pareto-set, determining a distribution over rankings for the objects in the set using the decision space, where a proportion is associated with each ranking in the distribution; selecting a sequence of rankings for the objects in the set based on the distribution in accordance with their proportions; and outputting the selected sequence of rankings of the objects.

Ranking systems and methods using a dynamic bayesian network exposure model

A computer-implemented method of locating points of interest for a user in a geographic area is disclosed. A current state of the user is used as an index into a list of a plurality of current state functions to select at least one of the plurality of current state functions. Each current state function corresponds with a utility measure set where each utility measure set includes a plurality of utility measures ordered in a selected sequence. The computer is used to perform: (i) lexicographic optimization with respect to the plurality of current state functions so that the selected sequence of each utility measure set is optimally ordered, and (ii) rate a plurality of points of interest with respect to at least one of the plurality of optimized current state functions, and, based on the rating, to automatically select a set of points of interest for the user.

Method of locating points of interest in a geographic area

A system and method perform cross-modal information retrieval, by generating a graph representing the set of media objects. Each node of the graph corresponds to a media object and is labeled with a set of features corresponding to a text part of the respective media object. Each edge between two nodes represents a similarity between a media part of the two nodes. A first relevance score is computed for each media object of the set of media objects that corresponds to a text-based score. A second relevance score is computed for each media object by inputting the graph into a graph neural network. The first relevance score and the second relevance score are combined to obtain a final ranking score for each media object.

System and method for performing cross-modal information retrieval using a neural network using learned rank images

A dialog device comprises a natural language interfacing device (chat interface or a telephonic device), and a natural language output device (the chat interface, a display device, or a speech synthesizer outputting to the telephonic device). A computer stores natural language dialog conducted via the interfacing device and constructs a current utterance word-by-word. Each word is chosen by applying a plurality of language models to a context comprising concatenation of the stored dialog and the current utterance thus far. Each language model outputs a distribution over the words of a vocabulary. A recurrent neural network (RNN) is applied to the distributions to generate a mixture distribution. The next word is chosen using the mixture distribution. The output device outputs the current natural language utterance after it has been constructed by the computer.

Dialog device with dialog support generated using a mixture of language models combined using a recurrent neural network

A system and method are disclosed which enable more effective email response authoring by contact center agents, for example, by automatically suggesting prototypical (entire) email responses to the human agent and interactive suggestion of next sentence candidates during the writing process. In one method, a customer inquiry is received and a latent topic prediction is generated, based on a word-based representation of the customer inquiry. A latent topic prediction is generated for an entire agent's reply to the customer inquiry as a function of the latent topic prediction generated for the customer inquiry. A further latent topic prediction is generated for a next sentence of the agent's reply as a function of a topic prediction for the next sentence which is generated with a prediction model that has been trained on annotated sentences of agent replies. Information is output to assist the agent, based on the topic predictions.

Method and system for assisting contact center agents in composing electronic mail replies

A method, system, and computer program product for extracting text motifs from the electronic documents is disclosed. A user provides a largest-maximal repeat or a super-maximal repeat as a first text block. The occurrences of the first text block are detected to identify the second text blocks in the vicinity of the occurrences of the first text block on the basis of pre-defined parameters. The text motifs are determined by combining the first text block and the second text block. Finally, the text motifs are extracted from the electronic documents.

Method and system for motif extraction in electronic documents

A computer implemented tactile user interface (TUI) and a method of manipulating objects with a virtual magnet are provided. The TUI includes a display comprising a touch-screen. The display is configured for displaying a set of graphic objects, each graphic object representing a respective one of a set of items, such as documents, e.g., text documents or images. A virtual magnet is caused to move on the display, in response to touching on the touch-screen, e.g., by dragging a finger or other implement across. The magnet is associated with a particular function command such that a subset of the graphic objects exhibits a response to the virtual magnet (e.g., is caused to move, relative to the virtual magnet or exhibits another visible response), each graphic object in the subset moving or otherwise responding as a function of an attribute of the underlying item represented by the graphic object.

Manipulation of displayed objects by virtual magnetism

A retrieval method on a database of documents including text and names of participants associated with the documents includes: receiving a text query facet of keywords and a persons query facet of participant names; computing an enriched text query as an aggregation of the text query facet, a monomodal expansion of the text query facet based on the keywords, a cross-modal expansion of the text query facet based on the participant names, and a topic expansion of the text query facet based on a topic model associating words and topics; computing an enriched persons query as an aggregation of the persons query facet, a monomodal expansion of the persons query facet based on the participant names, a cross-modal expansion of the persons query facet based on the keywords, and a community expansion of the persons query facet based on a community model associating persons and communities.

Mailbox search engine using query multi-modal expansion and community-based smoothing

A method and a system for coreference resolution are provided. The method includes receiving a set of document clusters, each cluster in the set of document clusters including a set of text documents. Instances of each of a set of candidate named entities are identified in the document clusters. For a pairs of the candidate named entities, at least one socio-temporal feature is computed that is based on the similarity of the distributions of identified instances of the respective candidate name entities among the document clusters. A decision for merging for the candidate named entities into a common real named entity is based on the socio-temporal features.

System and method for resolving entity coreference

A multimedia information retrieval system includes a storage and an electronic processing device. The latter is configured to perform a process including: computing values of a pairwise similarity measure quantifying pairwise similarity of documents of a multimedia reference repository; storing the computed values in the storage; performing an initial information retrieval process respective to the multimedia reference repository to return a set of initial repository documents; and identifying a set of top ranked documents of the multimedia reference repository based at least on the stored computed values pertaining to the set of initial repository documents.

Cross-media similarity measures through trans-media pseudo-relevance feedback and document reranking

A computer implemented system and method are provided for refining category scores for pages of a sequence of document pages that potentially includes document boundaries. The method uses initial category scores provided by a categorizer that considers one page at a time or concatenated pairs of pages (called bipages). The category scores represent the probability that a page belongs to a particular category. The method uses anisotropic diffusion to refine the initial page category scores using the scores of neighboring pages as a function of the probability that there is a boundary between the pages. The method may be performed iteratively.

Categorization of multi-page documents by anisotropic diffusion

A method for clustering documents is provided. Each document is represented by a multidimensional data point. The data points are initially assigned to a respective cluster and serve as their initial representative points. Thereafter, in an iterative process, the data points are clustered among the clusters, by assigning the data points to the clusters based on a comparison measure of each data point with the cluster or its representative point, and a threshold of the comparison measure. Based on this clustering, a new representative point for each of the clusters can be computed. Optionally, overlapping clusters are merged. For the next iteration, the new representative points are used as the representative points. An assignment of the documents to the clusters is output, based on a clustering of the data points in the latest iteration. Multiple batches may be processed, retaining the initial clusters to which the original batch was assigned.

Full and semi-batch clustering

A set of documents is annotated by metadata specifying persons associated with documents and their social roles in the documents. The annotated documents define a group of representation modes including at least one content type and at least one social role. An electronic processing device computes a relevance score for a person of interest using a set of queries each having a target social role by performing a sequence of operations that includes the following operations: computing similarities between documents and queries with respect to at least one similarity mode of the group of representation modes; enriching queries or documents to identify and aggregate nearest neighbor documents that are most similar with respect to at least one enrichment mode of the group of representation modes; aggregating over documents; aggregating over queries; and aggregating over at least one of (i) enrichment modes, (ii) similarity modes, and (iii) target social roles.

Growing community of inventors

Quaix-en-Chartreuse, France

Jean-Michel Renders