Lexical Vector Research Articles

Research on Characterization and Shaping of Chinese Heroic Characters in Online Media Based on Deep Neural Network Evaluation

Shaping the image of heroic characters is an important task of melodramatic films and TV dramas, whether they are revolutionary heroes, combat heroes, leader heroes, or civilian heroes, which have always been the key objects of melodramatic films and TV dramas. However, the traditional melodramatic films and TV dramas are facing many bottlenecks in the acceptance of heroic characters in the new era. In this paper, in order to obtain a richer character text, we divide the denotation elimination into two parts; one is the denotation elimination without considering the zero pronoun phenomenon, which is mainly done by the existing end-to-end neural codenotation algorithm, and the other is the attention-based algorithm model for the zero pronoun phenomenon. Analyzing this result from the perspective of the algorithmic model, it can be speculated that the lexical vector has a superior performance in the overall predicted character score, indicating that in the Big Five personality analysis for characters, the character’s personality is more reflected in lexicality, and the scores on a person’s Big Five personality dimensions can be indirectly reflected by the frequency of using different lexicalities. The similarities and differences among heroes are explored to create a popular hero image.

Применение методов искусственного интеллекта для решения задач поиска семантических ассоциатов на примере топонима «Москва»

Actual problems of toponymy imply the study of individual words in order to restore the conceptual meaning of geographical names lost, to find out how they reflected the characteristic features of the terrain, the type of ac-tivity of the people inhabiting it, etc. The purpose of the study is to determine the origin of the toponym Moskva by using artificial intelligence methods. The GeoWAC fastText embedding model based on the corpus of Russian-language texts of the RusVecteres service is used to calculate semantic similarity between words. The model assumes defining the semantic associates of toponyms by using the vector representation of words in the semantic space and finding the lexical vectors most closely located to the vector of the original word. To analyze a toponym there is applied a methods of semantic associates, a cluster analysis, a combined method based on the method of transformation of a word with a lost meaning and the analysis of semantic associates for a set of word transformants. The method is formalized by using a model that determines the similarity of the studied word and associates based on different versions of the model for one or more text corpora. The associated words obtained by the artificial intelligence are considered as a semantic cluster, and the calculated cosine similarity between vectors is considered as a measure of the similarity of elements in the cluster. To identify various hypotheses of the origin of the toponym Moskva there has been carried out a cluster analysis of the totality of the first ten vector associates for all transformants of this word. As a result, four hypotheses were advanced: “a famous man”, “firearms”, “beekeeping”, “blood-sucking insects”. The probabilities of the occurrence of these hypotheses are based on the study of the frequency of words in the corpus 
 of the language. The main hypothesis is a “famous person”.

The quest for better clinical word vectors: Ontology based and lexical vector augmentation versus clinical contextual embeddings

BackgroundWord vectors or word embeddings are n-dimensional representations of words and form the backbone of Natural Language Processing of textual data. This research experiments with algorithms that augment word vectors with lexical constraints that are popular in NLP research and clinical domain constraints derived from the Unified Medical Language System (UMLS). It also compares the performance of the augmented vectors with Bio + Clinical BERT vectors which have been trained and fine-tuned on clinical datasets. MethodsWord2vec vectors are generated for words in a publicly available de-identified Electronic Health Records (EHR) dataset and augmented by ontologies using three algorithms that have fundamentally different approaches to vector augmentation. The augmented vectors are then evaluated alongside publicly available Bio + Clinical BERT on their correlation with human-annotated lists using Spearman's correlation coefficient. They are also evaluated on the downstream task of Named Entity Recognition (NER). Quantitative and empirical evaluations are used to highlight the strengths and weaknesses of the different approaches. ResultsThe counter-fitted word2vec vectors augmented with information from the UMLS ontology produced the best correlation overall with human-annotated evaluation lists (Spearman's correlation of 0.733 with mini mayo-doctors’ annotation) while Bio + Clinical BERT produces the best results in the NER task (F1 of 0.87 and 0.811 on the i2b2 2010 and i2b2 2012 datasets respectively) in our experiments. ConclusionClinically adapted word2vec vectors successfully encapsulate concepts of lexical and clinical synonymy and antonymy and to a smaller extent, hyponymy and hypernymy. Bio + Clinical BERT vectors perform better at NER and avoid out-of-vocabulary words.

Optimizing biomedical ontology alignment in lexical vector space

Optimizing biomedical ontology alignment in lexical vector space

Fracture Mechanics Method for Word Embedding Generation of Neural Probabilistic Linguistic Model

Word embedding, a lexical vector representation generated via the neural linguistic model (NLM), is empirically demonstrated to be appropriate for improvement of the performance of traditional language model. However, the supreme dimensionality that is inherent in NLM contributes to the problems of hyperparameters and long-time training in modeling. Here, we propose a force-directed method to improve such problems for simplifying the generation of word embedding. In this framework, each word is assumed as a point in the real world; thus it can approximately simulate the physical movement following certain mechanics. To simulate the variation of meaning in phrases, we use the fracture mechanics to do the formation and breakdown of meaning combined by a 2-gram word group. With the experiments on the natural linguistic tasks of part-of-speech tagging, named entity recognition and semantic role labeling, the result demonstrated that the 2-dimensional word embedding can rival the word embeddings generated by classic NLMs, in terms of accuracy, recall, and text visualization.