N-gram Language Models Research Articles

Automatic Semantic Annotation of Indonesian Language Phrase Using N-Gram Language Model

Building semantic data populations in unstructured data or text is challenging. In this type of data, several problems can be raised, some of which are difficult to analyze. Some groups of words or expressions cannot be defined according to their meaning and can be a source of ambiguity. It can have a different meaning depending on the context of its use. This work aims to automatically annotate Indonesian Language text, especially phrases, with the existing knowledge base. The result is text with semantic markup. Machines can automatically process this type of text because it describes its meaning. This work applies an n-gram language model to identify meaningful phrases and defines them as a unit so that every existing word or phrase is automatically semantically tagged. This work uses the DBpedia and schema.org knowledge base. The percentage of successfully labeled data in this job was 78% with 84.95% accuracy using DBpedia and 5.9% with 97.46% accuracy using schema .org. Some factors affect the accuracy score, including the availability of the required data with the data contained in the knowledge base, the system's ability in the POS tagging process, and many new terminology and local cultures that have not yet been contained in the knowledge bases, especially schema.org that is utilized as a standard for all search engines. This work will help the machine understand the semantics of text data. All pages obtained will be semantically tagged and, therefore, will be understood by machines. This ability will support the following processes.

English grammar intelligent error correction technology based on the n-gram language model

Abstract With the development of the Internet, the number of electronic texts has increased rapidly. Automatic grammar error correction technology is an effective safeguard measure for the quality of electronic texts. To improve the quality of electronic text, this study introduces a moving window algorithm and linear interpolation smoothing algorithm to build a Cn-gram language model. On this basis, a syntactic analysis strategy is introduced to construct a syntactic error correction model integrating Cn-gram and syntactic analysis, and English grammar intelligent error correction is carried out through the model. The results show that compared with the Bi-gram and Tri-gram, the precision of the Cn-gram model is 0.85 and 0.91% higher, and the F1 value is 0.97 and 1.14% higher, respectively. Compared with the results of test set Long, the Cn-gram model has better performance on verb error correction of the Short test set, and the precision rate, recall rate, and F1 value are increased by 0.86, 3.94, and 1.87%, respectively. The comparison of the precision, recall rate, and F1 value of the proposed grammar error correction model on the complete test set shows that the precision of the study is 19.10 and 5.41% higher for subject–verb agreement errors. The recall rate is 9.55 and 10.77% higher, respectively; F1 values are higher by 12.65 and 10.59%, respectively. The above results show that the error-correcting technique of the research design has excellent error-correcting performance. It is hoped that this experiment can provide a reference for the relevant research of automatic error correction technology of electronic text.

Arithmetic N-gram: an efficient data compression technique

Due to the increase in the growth of data in this era of the digital world and limited resources, there is a need for more efficient data compression techniques for storing and transmitting data. Data compression can significantly reduce the amount of storage space and transmission time to store and transmit given data. More specifically, text compression has got more attention for effectively managing and processing data due to the increased use of the internet, digital devices, data transfer, etc. Over the years, various algorithms have been used for text compression such as Huffman coding, Lempel-Ziv-Welch (LZW) coding, arithmetic coding, etc. However, these methods have a limited compression ratio specifically for data storage applications where a considerable amount of data must be compressed to use storage resources efficiently. They consider individual characters to compress data. It can be more advantageous to consider words or sequences of words rather than individual characters to get a better compression ratio. Compressing individual characters results in a sizeable compressed representation due to their less repetition and structure in the data. In this paper, we proposed the ArthNgram model, in which the N-gram language model coupled with arithmetic coding is used to compress data more efficiently for data storage applications. The performance of the proposed model is evaluated based on compression ratio and compression speed. Results show that the proposed model performs better than traditional techniques.

A search tool based on language modelling developed for The Index of Middle English Prose

Non-standardised early vernaculars present a problem for search tools due to the high degree of variation. The challenge lies in the variation found in orthography, syntax, and lexicon between titles, incipits, and explicits in manuscript copies of the same work. Traditional search methods relying on exact string matching or regular expressions fail to address these variations comprehensively. This project presents a web-based search tool specifically designed to handle linguistic and textual variation. The software is made available as a part of the Index of Middle English Prose (IMEP). The search tool addresses the issue of variation by utilizing a database of incipits and explicits, character-based n-gram language models (LMs) built with the Stanford Research Institute Language Modelling (SRILM) toolkit, and a fuzzy search script (IMEP: FSS) written in Python. The tool optimizes for recall, retrieving multiple potential matches for a search string, without attempting to identify the ‘correct’ one. The search process involves looking up exact matches in the database while simultaneously using the fuzzy search script to evaluate the incipits and explicits against a model of the search string, followed by a match of the search string against models of the incipits and explicits. This two-step process shortens the processing time, which would otherwise be unreasonably long, because while using SRILM to match the search string against each incipit or explicit in the IMEP for precision could be time-consuming, running a first step where all texts are matched against a single LM built from the search string allows for faster processing. A web application, built using Django and Docker, combines the results of the direct database lookup and the fuzzy search script, presenting them as a list with exact matches followed by fuzzy matches ordered by increasing model perplexity. The tool is made available Open Access and can be adapted to other datasets.

N-gram Language Model for Chinese Function-word-centered Patterns

N-gram language modelling, a proven and effective method in NLP, is widely used to calculate the probability of a sentence in natural language. Language pattern is a linguistic level between word/character and sentence, which exists in pattern grammar. In this research, the approach of language model and language pattern are combined for the first time, and language patterns are studied by use of the N-gram model. Chinese function-word-centered patterns are extracted from the LCMC corpus, and aligned into pattern chains. The language model is trained from these chains to investigate the properties and distribution of Chinese function words, the interaction of content words and function words, and the interaction between patterns. The results indicate that there are approximately 10,000 function-word-centered patterns in the texts, which are distributed exponentially. This research summarizes the most common function-word-centered patterns and content-word-centered patterns, and discusses the interactions of patterns based on corpus data. The bigram language model of these patterns reflects the restrictions of function words. In addition, the research adopts an innovative method to visualize the interactions between patterns. This research fills the research gap between word/character and sentence and reveals basic Chinese pattern categories and the interactions between patterns, which makes a significant contribution to Chinese linguistic research, and improves the efficiency of NLP.

Speech Recognition Enhancement and Compression Perception in Russian Translation Teaching Cooperative System Application

Abstract In the digital age, characterized by the ubiquity of big data and artificial intelligence, Russian translation education is poised for transformative change. This study explores the integration of advanced speech recognition and compressed perception technologies into Russian translation instruction, highlighting their potential to elevate educational quality and translator training. Utilizing the Spring Boot framework, we developed a comprehensive Russian translation teaching system, incorporating a Russian-Chinese speech translation model that leverages the RNN Transducer model and an N-gram language model, alongside compressed perception theory for efficient speech signal processing. Our findings reveal a notable enhancement in Russian speech recognition accuracy and processing speed, with a 4.73% improvement in endto-end speech detection accuracy and a significant reduction in word error rate, outperforming CTC and CTC-LM models by 14.79% to 24.85%. Furthermore, speech quality and intelligibility scores experienced marked improvements. The application of speech recognition enhancement and compressed perception technologies emerges as a vital strategy for enriching Russian translation pedagogy and fostering the development of proficient translators.

Intelligent Translation System Aiding High-Quality Writing in English in the Age of the Internet

Abstract This study focuses on applying intelligent translation systems in assisting high-quality English writing in the Internet era. The study analyzes statistical machine translation techniques, especially N-gram language modeling and word alignment techniques, and their crucial role in translation quality improvement. In English writing, the intelligent translation system significantly improves the quality of students’ Writing through word block translation. By analyzing 320 students, the mean value of self-efficacy in writing skills of students in the high-quality writing group was 3.78, significantly higher than that of the low-quality group, which was 2.82. After the experiment, 0.57 students indicated that they would improve their English writing vocabulary with the aid of the Intelligent Translation System, which showed the potential of the Intelligent Translation System to enhance students’ interest in and autonomy in Writing. Average distribution analysis shows that word block usage positively correlates with writing performance, with an R Square value of 0.6726. The intelligent translation system improves students’ English writing and enhances their self-efficacy, which is of great significance to English teaching.

Research on the Influence of Socialist Core Value System Construction on College Students’ Ideological and Political Education in the Context of Deep Learning

Abstract In this paper, the word-level N-gram language model applicable to Chinese word segmentation in ideological and political education is obtained through large-scale corpus training, and the Chinese word segmentation path in ideological and political education is calculated according to the conditional probability formula. On the basis of the deep learning algorithm, the Chinese participle algorithm is optimized by introducing word frequency deviation and sorted word frequency deviation indexes, and the cut score with the largest value of the participle indexes is taken as the final participle result output. Starting from the current situation, we determine the evaluation indexes of ideological and political education. Meanwhile, we complete the research design of the impact of the socialist core value system and use the SPSS22.0 and Origin2019 statistical software to carry out the research and analysis of ideological and political education. The results show that the index values of the Chinese word separation algorithm based on the 4-gram language model have also been improved by 1.15%, which means that there exists the possibility of practical application of the model in Chinese word separation in ideological and political education. The college students’ identification with the socialist core values at the national, social and personal levels are all at a high level, with values of 4.28±0.82, 4.32±0.75 and 4.61±0.62 respectively. This study improves students’ ideological-political literacy and provides references to their healthy growth.

A search tool based on language modelling developed for The Index of Middle English Prose.

Non-standardised early vernaculars present a problem for search tools due to the high degree of variation. The challenge lies in the variation found in orthography, syntax, and lexicon between titles, incipits, and explicits in manuscript copies of the same work. Traditional search methods relying on exact string matching or regular expressions fail to address these variations comprehensively. This project presents a web-based search tool specifically designed to handle linguistic and textual variation. The software is made available as a part of the Index of Middle English Prose (IMEP). The search tool addresses the issue of variation by utilizing a database of incipits and explicits, character-based n-gram language models (LMs) built with the Stanford Research Institute Language Modelling (SRILM) toolkit, and a fuzzy search script (IMEP: FSS) written in Python. The tool optimizes for recall, retrieving multiple potential matches for a search string, without attempting to identify the 'correct' one. The search process involves looking up exact matches in the database while simultaneously using the fuzzy search script to evaluate the incipits and explicits against a model of the search string, followed by a match of the search string against models of the incipits and explicits. This two-step process shortens the processing time, which would otherwise be unreasonably long, because while using SRILM to match the search string against each incipit or explicit in the IMEP for precision could be time-consuming, running a first step where all texts are matched against a single LM built from the search string allows for faster processing. A web application, built using Django and Docker, combines the results of the direct database lookup and the fuzzy search script, presenting them as a list with exact matches followed by fuzzy matches ordered by increasing model perplexity. The tool is made available Open Access and can be adapted to other datasets.

Improving speech recognition systems for the morphologically complex Malayalam language using subword tokens for language modeling

This article presents the research work on improving speech recognition systems for the morphologically complex Malayalam language using subword tokens for language modeling. The speech recognition system is built using a deep neural network–hidden Markov model (DNN-HMM)-based automatic speech recognition (ASR). We propose a novel method, syllable-byte pair encoding (S-BPE), that combines linguistically informed syllable tokenization with the data-driven tokenization method of byte pair encoding (BPE). The proposed method ensures words are always segmented at valid pronunciation boundaries. On a text corpus that has been divided into tokens using the proposed method, we construct statistical n-gram language models and assess the modeling effectiveness in terms of both information-theoretic and corpus linguistic metrics. A comparative study of the proposed method with other data-driven (BPE, Morfessor, and Unigram), linguistic (Syllable), and baseline (Word) tokenization algorithms is also presented. Pronunciation lexicons of subword tokenized units are built with pronunciation described as graphemes. We develop ASR systems employing the subword tokenized language models and pronunciation lexicons. The resulting ASR models are comprehensively evaluated to answer the research questions regarding the impact of subword tokenization algorithms on language modeling complexity and on ASR performance. Our study highlights the strong performance of the hybrid S-BPE tokens, achieving a notable 10.6% word error rate (WER), which represents a substantial 16.8% improvement over the baseline word-level ASR system. The ablation study has revealed that the performance of S-BPE segmentation, which initially underperformed compared to syllable tokens with lower amounts of textual data for language modeling, exhibited steady improvement with the increase in LM training data. The extensive ablation study indicates that there is a limited advantage in raising the n-gram order of the language model beyond n=3. Such an increase results in considerable model size growth without significant improvements in WER. The implementation of the algorithm and all associated experiments are available under an open license, allowing for reproduction, adaptation, and reuse.

Training RNN language models on uncertain ASR hypotheses in limited data scenarios

Training domain-specific automatic speech recognition (ASR) systems requires a suitable amount of data comprising the target domain. In several scenarios, such as early development stages, privacy-critical applications, or under-resourced languages, only a limited amount of in-domain speech data and an even smaller amount of manual text transcriptions, if any, are available. This motivates the study of ASR language model (LM) training on a limited amount of in-domain speech data. Early works have attempted training of n-gram LMs from ASR N-best lists and lattices but training and adaptation of recurrent neural network (RNN) LMs from ASR transcripts has not received attention. In this work, we study training and adaptation of RNN LMs using alternate, uncertain ASR hypotheses embedded in ASR confusion networks obtained from target domain speech data. We explore different methods for training the RNN LMs to deal with the uncertain input sequences. The first method extends the cross-entropy objective into a Kullback–Leibler (KL) divergence based training loss, the second method formulates a training loss based on a hidden Markov model (HMM), and the third method performs training on paths sampled from the confusion networks. These methods are applied to limited data setups including telephone and meeting conversation datasets. Performance is evaluated under two settings wherein no manual transcriptions or a small amount of manual transcriptions are available to aid the training. Moreover, a model adaptation setting is also evaluated wherein the RNN LM is pre-trained on an out-of-domain conversational corpus. Overall the sampling method for training RNN LMs on ASR confusion networks performs the best, and results in up to 12% relative reduction in perplexity on the meeting dataset as compared to training on ASR 1-best hypotheses, without any manual transcriptions. However, the perplexity reductions do not translate into equivalent WER reductions. A detailed analysis of the perplexity reductions obtained by the different methods is performed in order to understand this effect.

High-order interaction feature selection for classification learning: A robust knowledge metric perspective

Feature selection is an important learning task in data mining and knowledge discovery. Nevertheless, the fuzziness, uncertainty, and noise presented by the data greatly complicate the construction of learning models. Moreover, most works focus on exploring low-order correlations between variables using low-dimensional mutual information, without paying attention to high-order interaction for multiple variables, resulting in the loss of some potentially important dependency information. Driven by these two issues, a robust knowledge metric approach is invented to perceive and excavate the latent information hidden in interaction. In this study, firstly, a robust fuzzy granularity space is constructed from different granular structures induced by different features, and the robust fuzzy uncertainty measures (RFUMs) are successively devised. Then, RFUMs are used to measure pair-wise, three-order, and even higher-order interaction dependencies among features. Further, a constrained high-order interaction evaluation function inspired by the N-gram language model is formulated, and a corresponding high-order interaction feature selection algorithm with RFUMs (HIFS-RFUMs) is designed. Next, comparative experiments with seven representative algorithms on twenty datasets illustrate its effectiveness. In addition, ablation experiments are conducted on the high-order interaction feature selection algorithm with fuzzy uncertainty measures (HIFS-FUMs) and the relative reduction algorithm with RFUMs (R2-RFUMs), which demonstrate the robustness of the metric and the effectiveness for mining high-order interactive features, respectively.

Self-correction of automatic speech recognition

The rapid development of modern technology and society has raised the need for a large number of real-time translations, such as Automatic Speech Recognition (ASR) technology. One of the important factors determining the fluency of cooperation is that ASR translates the input speech into readable text so that both parties can understand each other's meaning. There is also a huge demand for and reliance on automatic speech recognition technology for modern applications. This article focuses on the word matching algorithm for backward search (WMA-BS), which requires two separate steps, one for the word matching algorithm and the other for the backward search step, for the first step, which can also be done by applying the word-level N-Gram language model. n-Gram model is a language model in which when n words appear in a text, we can use these n words to predict the structure of the text. we can use these n words to predict the structure of the text. In order to make the word-level N-gram language model suitable for the method, it is more useful for us to integrate it into shallow fusion. One of the most important parts to be addressed in the word matching algorithm is the boundary, or distance, between two words in English.

An information theoretic score for learning hierarchical concepts.

How do humans learn the regularities of their complex noisy world in a robust manner? There is ample evidence that much of this learning and development occurs in an unsupervised fashion via interactions with the environment. Both the structure of the world as well as the brain appear hierarchical in a number of ways, and structured hierarchical representations offer potential benefits for efficient learning and organization of knowledge, such as concepts (patterns) sharing parts (subpatterns), and for providing a foundation for symbolic computation and language. A major question arises: what drives the processes behind acquiring such hierarchical spatiotemporal concepts? We posit that the goal of advancing one's predictions is a major driver for learning such hierarchies and introduce an information-theoretic score that shows promise in guiding the processes, and, in particular, motivating the learner to build larger concepts. We have been exploring the challenges of building an integrated learning and developing system within the framework of prediction games, wherein concepts serve as (1) predictors, (2) targets of prediction, and (3) building blocks for future higher-level concepts. Our current implementation works on raw text: it begins at a low level, such as characters, which are the hardwired or primitive concepts, and grows its vocabulary of networked hierarchical concepts over time. Concepts are strings or n-grams in our current realization, but we hope to relax this limitation, e.g., to a larger subclass of finite automata. After an overview of the current system, we focus on the score, named CORE. CORE is based on comparing the prediction performance of the system with a simple baseline system that is limited to predicting with the primitives. CORE incorporates a tradeoff between how strongly a concept is predicted (or how well it fits its context, i.e., nearby predicted concepts) vs. how well it matches the (ground) "reality," i.e., the lowest level observations (the characters in the input episode). CORE is applicable to generative models such as probabilistic finite state machines (beyond strings). We highlight a few properties of CORE with examples. The learning is scalable and open-ended. For instance, thousands of concepts are learned after hundreds of thousands of episodes. We give examples of what is learned, and we also empirically compare with transformer neural networks and n-gram language models to situate the current implementation with respect to state-of-the-art and to further illustrate the similarities and differences with existing techniques. We touch on a variety of challenges and promising future directions in advancing the approach, in particular, the challenge of learning concepts with a more sophisticated structure.

Kurdish Kurmanji Lemmatization and Spell-checker with Spell-correction

There are many studies about using lemmatization and spell-checker with spell-correction regarding English, Arabic, and Persian languages but only few studies found regarding low-resource languages such as Kurdish language and more specifically for Kurmanji dialect, which increased the need of creating such systems. Lemmatization is the process of determining a base or dictionary form (lemma) for a specific surface pattern, whereas spell-checkers and spell-correctors determine whether a word is correctly spelled also correct a range of spelling errors, respectively. This research aims to present a lemmatization and a word-level error correction system for Kurdish Kurmanji Dialect, which are the first tools for this dialect based on our knowledge. The proposed approach for lemmatization is built on morphological rules, and a hybrid approach that relies on the n-gram language model and the Jaccard Coefficient Similarity algorithm was applied to the spell-checker and spell-correction. The process results for lemmatization, as detailed in this article, rates of 97.7% and 99.3% accuracy for noun and verb lemmatization, correspondingly. Furthermore, for spell-checker and spell-correction, accordingly, accuracy rates of 100% and 90.77% are attained.

An intelligent extension of the training set for the Persian n-gram language model: an enrichment algorithm

In this article, we are going to introduce an automatic mechanism to intelligently extend the training set to improve the n-gram language model of Persian. Given the free word-order property in Persian, our enrichment algorithm diversifies n-gram combinations in baseline training data through dependency reordering, adding permissible sentences and filtering ungrammatical sentences using a hybrid empirical (heuristic) and linguistic approach. Experiments performed on baseline training set (taken from a standard Persian corpus) and the resulting enriched training set indicate a declining trend in average relative perplexity (between 34% to 73%) for informal/spoken vs. formal/written Persian test data.

An End-to-End Transformer-Based Automatic Speech Recognition for Qur’an Reciters

The attention-based encoder-decoder technique, known as the trans-former, is used to enhance the performance of end-to-end automatic speech recognition (ASR). This research focuses on applying ASR end-to-end transformer-based models for the Arabic language, as the researchers’ community pays little attention to it. The Muslims Holy Qur’an book is written using Arabic diacritized text. In this paper, an end-to-end transformer model to building a robust Qur’an vs. recognition is proposed. The acoustic model was built using the transformer-based model as deep learning by the PyTorch framework. A multi-head attention mechanism is utilized to represent the encoder and decoder in the acoustic model. A Mel filter bank is used for feature extraction. To build a language model (LM), the Recurrent Neural Network (RNN) and Long short-term memory (LSTM) were used to train an n-gram word-based LM. As a part of this research, a new dataset of Qur’an verses and their associated transcripts were collected and processed for training and evaluating the proposed model, consisting of 10 h of .wav recitations performed by 60 reciters. The experimental results showed that the proposed end-to-end transformer-based model achieved a significant low character error rate (CER) of 1.98% and a word error rate (WER) of 6.16%. We have achieved state-of-the-art end-to-end transformer-based recognition for Qur’an reciters.

Forward-backward Transliteration of Punjabi Gurmukhi Script Using N-gram Language Model

Transliterating the text of a language to a foreign script is called forward transliteration and transliterating the text back to the original script is called backward transliteration. In this work, we perform both forward as well as backward transliteration on Punjabi. We transliterate Punjabi person names from Gurmukhi script to English Roman script and from English Roman script back to Gurmukhi script using n-gram language model. We used more than one million parallel entities of person names in Gurmukhi and Roman script as the training corpus. We generated English to Punjabi and Punjabi to English n-grams databases from the corpus. To get better results, we tried to create as long n-grams as possible ranging from bi-gram to 30-gram. Our n-grams database contains more than 10 million n-grams, with each n-gram having multiple mappings of the other script. The most challenging part is to find the mapping for the given n-gram from the parallel name entity while creating n-grams databases. As per the orthography rules, the same combination of letters may have different pronunciation, depending upon its location in the word. Therefore, we categorized n-grams into starting, middle, and ending n-grams and used them accordingly in the transliteration process. The transliteration process works like the merge sort. We start searching the longest possible n-gram in the database and split the string recursively until the match is found. The transliterated strings are merged back to form the final output. In English to Punjabi transliteration, we achieved 96% accuracy using gold standard and 99.14% accuracy using minimum edit distance. In Punjabi to English transliteration, the result showed 96.85% and 99.35% accuracy for the gold standard and minimum edit distance, respectively.

Central Kurdish Automatic Speech Recognition using Deep Learning

Automatic Speech Recognition (ASR) as an interesting field of speech processing, is nowadays utilized in real applications which are implemented using various techniques. Amongst them, the artificial neural network is the most popular one. Increasing the performance and making these systems robust to noise are among the current challenges. This paper addresses the development of an ASR system for the Central Kurdish language (CKB) using a transfer learning of Deep Neural Networks (DNN). The combination of Mel-Frequency Cepstral Coefficients (MFCCs) for extracting features of speech signals, Long Short-Term Memory (LSTM) with Connectionist Temporal Classification (CTC) output layer is used to create an Acoustic Model (AM) on the AsoSoft CKB speech dataset. Also, we have used the N-gram language model on the collected large text dataset which includes about 300 million tokens. The text corpus is also used to extract a dynamic lexicon model that contains over 2.5 million CKB words. The obtained results show that the use of a DNN improves the results compared to classical statistics modules. The proposed method achieves a 0.22%-word error rate by combining transfer learning and language model adaptation. This result is superior to the best-reported result for the CKB.

Using a classification model for determining the value of liver radiological reports of patients with colorectal cancer.

Medical imaging is critical in clinical practice, and high value radiological reports can positively assist clinicians. However, there is a lack of methods for determining the value of reports. The purpose of this study was to establish an ensemble learning classification model using natural language processing (NLP) applied to the Chinese free text of radiological reports to determine their value for liver lesion detection in patients with colorectal cancer (CRC). Radiological reports of upper abdominal computed tomography (CT) and magnetic resonance imaging (MRI) were divided into five categories according to the results of liver lesion detection in patients with CRC. The NLP methods including word segmentation, stop word removal, and n-gram language model establishment were applied for each dataset. Then, a word-bag model was built, high-frequency words were selected as features, and an ensemble learning classification model was constructed. Several machine learning methods were applied, including logistic regression (LR), random forest (RF), and so on. We compared the accuracy between priori choosing pertinent word strings and our machine language methodologies. The dataset of 2790 patients included CT without contrast (10.2%), CT with/without contrast (73.3%), MRI without contrast (1.8%), and MRI with/without contrast (14.6%). The ensemble learning classification model determined the value of reports effectively, reaching 95.91% in the CT with/without contrast dataset using XGBoost. The logistic regression, random forest, and support vector machine also achieved good classification accuracy, reaching 95.89%, 95.04%, and 95.00% respectively. The results of XGBoost were visualized using a confusion matrix. The numbers of errors in categories I, II and V were very small. ELI5 was used to select important words for each category. Words such as "no abnormality", "suggest", "fatty liver", and "transfer" showed a relatively large degree of positive correlation with classification accuracy. The accuracy based on string pattern search method model was lower than that of machine learning. The learning classification model based on NLP was an effective tool for determining the value of radiological reports focused on liver lesions. The study made it possible to analyze the value of medical imaging examinations on a large scale.