Abstract: The present disclosure provides a text error-correcting method, apparatus, electronic device and readable storage medium and relates to the field of natural language processing and deep learning. In the present disclosure, an implementation solution employed when performing text error correction is: obtaining a text to be processed, and an error-correcting type of the text to be processed; selecting a target error-correcting model corresponding to the error-correcting type; processing the text to be processed using the target error-correcting model, and regarding a processing result as an error-correcting result of the text to be processed. The present disclosure can enhance the flexibility and accuracy of text error correction.

Abstract: A pre-training method of a neural network model, an electronic device, and a medium. The pre-training data is inputted to the initial neural network model, and the initial neural network model is pre-trained in the first training mode, in the first training mode, the plurality of hidden layers share one hidden layer parameter, and the loss value of the initial neural network model is obtained, if the loss value of the initial neural network model is less than a preset threshold, the initial neural network model continues to be pre-trained in the second training mode, in the second training mode, each of the plurality of hidden layers has its own hidden layer parameter.

Abstract: The technical solutions relate to the fields of artificial intelligence technologies and natural language processing technologies. According to an embodiment, entity detection is performed on a query text to acquire a target entity; a feature representation of the query text is generated by using a pre-trained context representation model; and based on the feature representation of the query text and pre-acquired feature representations of entity categories corresponding to the target entity, the target entity is linked to an entity category with the highest matching degree.

Abstract: An active interaction method, an electronic device and a readable storage medium, relating to the field of deep learning and image processing technologies, are disclosed. According to an embodiment, the active interaction method includes: acquiring a video shot in real time; extracting a visual target from each image frame of the video, and generating a first feature vector of each visual target; for each image frame of the video, fusing the first feature vector of each visual target and identification information of the image frame to which the visual target belongs to generate a second feature vector of each visual target; aggregating the second feature vectors with the same identification information respectively to generate a third feature vector corresponding to each image frame; and initiating active interaction in response to determining that the active interaction is to be performed according to the third feature vector of a preset image frame.

Abstract: The present application discloses an optimizer learning method and apparatus, an electronic device and a readable storage medium, which relates to the field of deep learning technologies. An implementation solution adopted by the present application during optimizer learning is: acquiring training data, the training data including a plurality of data sets each including neural network attribute information, neural network optimizer information, and optimizer parameter information; and training a meta-learning model by taking the neural network attribute information and the neural network optimizer information in the data sets as input and taking the optimizer parameter information in the data sets as output, until the meta-learning model converges. The present application can implement self-adaptation of optimizers, so as to improve generalization capability of the optimizers.

Abstract: The disclosure discloses a method for generating a conversation, an electronic device, and a storage medium. The detailed implementation includes: obtaining a current conversation and historical conversations of the current conversation; selecting multiple reference historical conversations from the historical conversations and adding the multiple reference historical conversations to a temporary conversation set; and generating reply information of the current conversation based on the current conversation and the temporary conversation set.

Abstract: Embodiments of the present disclosure provide a method for generating a model for representing heterogeneous graph node. A specific implementation includes: acquiring a training data set, wherein the training data set includes node walk path information obtained by sampling a heterogeneous graph according to different meta paths; and training, based on a gradient descent algorithm, an initial heterogeneous graph node representation model with the training data set as an input of the initial heterogeneous graph node representation model, to obtain a heterogeneous graph node representation model.

Abstract: A method and apparatus for optimizing a recommendation system, a device and a computer storage medium are described, which relates to the technical field of deep learning and intelligent search in artificial intelligence. A specific implementation solution is: taking the recommendation system as an agent, a user as an environment, each recommended content of the recommendation system as an action of the agent, and a long-term behavioral revenue of the user as a reward of the environment; and optimizing to-be-optimized parameters in the recommendation system by reinforcement learning to maximize the reward of the environment. The present disclosure can effectively optimize long-term behavioral revenues of users.

Abstract: A method and apparatus for correcting character errors, an electronic device and a storage medium are disclosed, which relates to the natural language processing field and the deep learning field. The method may include: for a character to be processed, acquiring the score of each character in a pre-constructed vocabulary, the score being a score of the reasonability of the character in the vocabulary at the position of the character to be processed; selecting top K characters as candidates of the character to be processed, K being a positive integer greater than one; selecting an optimal candidate from the K candidates; and replacing the character to be processed with the optimal candidate if the optimal candidate is different from the character to be processed. With the solution of the present application, the accuracy of an error correction result, or the like, may be improved.

Abstract: The disclosure discloses a method for distilling a model, an electronic device, and a storage medium, and relates to the field of deep learning technologies. A teacher model and a student model are obtained. The second intermediate fully connected layer is transformed into an enlarged fully connected layer and a reduced fully connected layer based on a first data processing capacity of a first intermediate fully connected layer of the teacher model and a second data processing capacity of a second intermediate fully connected layer of the student model. The second intermediate fully connected layer is replaced with the enlarged fully connected layer and the reduced fully connected layer to generate a training student model. The training student model is distilled based on the teacher model.

Abstract: A method for dialogue processing, an electronic device and a storage medium are provided. The specific technical solution includes: obtaining a dialogue history; selecting a target machine from a plurality of machines; inputting the dialogue history into a trained dialogue model in the target machine to generate a response to the dialogue history, in which the dialogue model comprises a common parameter and a specific parameter, and different machines correspond to the same common parameter.

Abstract: Embodiments of the present disclosure provide a method and a device for predicting a box office trend of a film, a device and a storage medium. The method includes acquiring in real time a plurality of dynamic factor data of each of various films to be shown, in which, the dynamic factor data represents a factor that influences box office of the film; after a film in the various films is shown, incrementally updating a pre-trained box office prediction model by using box office data and the plurality of dynamic factor data of the film; and according to a preset period, predicting a box office trend of a target film to be predicted in the various films by using a box office prediction model incrementally updated in each preset period and the plurality of dynamic factor data of the target film, to obtain a plurality of prediction results.

Abstract: Embodiments of the present disclosure disclose a method, an apparatus, a server, and a storage medium for recalling for a search. The method for recalling for a search includes: acquiring a search term inputted by a user; calculating a semantic vector of the search term using a pre-trained neural network model; and recalling, according to a pre-established index, target documents related to the semantic vector of the search term from candidate documents, the index being established based on the semantic vectors of the candidate documents, and the semantic vectors of the candidate documents being calculated using the pre-trained neural network model. The embodiments of the present disclosure may solve a problem in the existing method for recalling that the recalling accuracy is affected by failing to generalize semantics, to improve the accuracy of recalling for a search.

Abstract: A method and device for correcting an error in a text are provided. The method includes: preprocessing the text to obtain at least one segment of the text; generating a plurality of candidate segments for the segment; scoring the plurality of candidate segments with a tree model, to obtain respective first scoring results of the plurality of candidate segments; scoring the plurality of candidate segments with a deep neural network model, to obtain respective second scoring results of the plurality of candidate segments; for each candidate segment, calculating a scoring of the candidate segment based on the first scoring result and the second scoring result of the candidate segment; ranking the plurality of candidate segments according to the scorings of the candidate segments, to obtain a ranking result; and correcting the error in the text according to the ranking result.

Abstract: A method for generating node representations in a heterogeneous graph, an electronic device, and a non-transitory computer-readable storage medium, and relates to the field of machine learning technologies. The method includes: acquiring a heterogeneous graph; inputting the heterogeneous graph into a heterogeneous graph learning model to generate a node representation of each node in the heterogeneous graph, in which the heterogeneous graph learning model generates the node representation of each node by actions of: segmenting the heterogeneous graph into a plurality of subgraphs, in which each subgraph includes nodes of two types and an edge of one type between the nodes of two types; and generating the node representation of each node according to the plurality of subgraphs.

Abstract: Embodiments of the present disclosure relate to a method and apparatus for generating a dialogue model. The method may include: acquiring a corpus sample set, a corpus sample including input information and target response information; classifying corpus samples in the corpus sample set, setting discrete hidden variables for the corpus samples based on a classification result to generate a training sample set, a training sample including the input information, the target response information, and a discrete hidden variable; and training a preset neural network using the training sample set to obtain the dialogue model, the dialogue model being used to represent a corresponding relationship between inputted input information and outputted target response information.

Abstract: Embodiments of the present disclosure provide a method and apparatus for processing data. The method may include: acquiring a sample set; inputting a plurality of target samples in the sample set into a pre-trained first natural language processing model, respectively, to obtain prediction results output from the pre-trained first natural language processing model; determining the obtained prediction results as labels of the target samples in the plurality of target samples, respectively; and training a to-be-trained second natural language processing model, based on the plurality of target samples and the labels of the target samples to obtain a trained second natural language processing model, parameters in the first natural language processing model being more than parameters in the second natural language processing model.

Abstract: Embodiment of the disclosure provide a method and apparatus for generating information. The method includes: acquiring vectors of a plurality of users, the vector being used to characterize behavior habits of the users; inputting the vectors of the plurality of users and push information pushed by a push system to the plurality of users into a feedback information generating model established in advance, to generate the feedback information of the plurality of users for the push information, wherein the feedback information generating model is used to characterize a corresponding relationship between the vectors, the push information and the feedback information; and generating an evaluation report of the push system based on the feedback information.

Abstract: A method and an apparatus for correcting a query based on artificial intelligence, including: receiving a first query input by a user, and judging whether the first query satisfies an error correcting condition according to a preset error correcting strategy; determining a first segment to be corrected in the first query if the first query satisfies the error correcting condition; acquiring one or more first candidate results corresponding to the first segment according to a preset candidate recalling strategy; determining an error correcting result corresponding to the first segment according to quality feature values of the one or more first candidate results; and performing an error correction on the first query according to the error correcting result, and generating a second query.

Abstract: Disclosed are a method and an apparatus for constructing a binary feature dictionary. The method may include: extracting binary features from a corpus; calculating a preset statistic of each binary feature; and selecting a preset number of binary features in sequence according to the preset statistic to constitute the binary feature dictionary.