Recent studies have demonstrated that supervised fine-tuning of Large Language Models (LLMs) can significantly enhance performance across various Information Extraction (IE) tasks. However, the critical IE task of Relation Extraction (RE) faces substantial cost barriers in the supervised fine-tuning of LLMs within few-shot learning contexts. Addressing this challenge, we introduce a hardware-efficient finetuning model for few-shot RE with large language model (LLMProto), which aims to fine-tune LLMs at low hardware costs, thereby improving the performance of few-shot RE tasks. LLMProto tackles few-shot RE by integrating the LLM Base Layer and Prototypical Network Layer.LLM Base Layer effectively reduces task complexity, and the Prototypical Network Layer captures underlying structural patterns in the data. Experimental results on the datasets demonstrate LLMProto’s superior performance in few-shot RE tasks, significantly outperforming existing baseline methods.

Chinese Spelling Correction (CSC) is a challenging and essential task in natural language processing. In this study, we introduces a new method for Chinese Spelling Correction (CSC) that addresses three unattended areas in prior studies. Firstly, we use an Implicit Knowledge Extraction Network to overcome limitations of conventional methods that rely on explicit knowledge alone. Secondly, we use KL divergence to limit the effect of incorrect characters on semantic understanding, ensuring consistent meaning. Finally, we employ a Cor-Det framework rather than the traditional Det-Cor framework, offering more consistent learning objectives. Tests on three SIGHAN benchmarks show this method significantly surpassing baseline models, highlighting the crucial role of implicit knowledge in Chinese Spelling Correction tasks.

Semantic text matching models have achieved outstanding performance, but traditional methods may not solve Few-shot learning problems and data augmentation techniques could suffer from semantic deviation. To solve this problem, we propose STMAP, which is implemented from the perspective of data augmentation based on Gaussian noise and Noise Mask signal. We also employ an adaptive optimization network to dynamically optimize the several training targets generated by data augmentation. We evaluated our model on four English datasets: MRPC, SciTail, SICK, and RTE, with achieved scores of 90.3%, 94.2%, 88.9%, and 68.8%, respectively. Our model obtained state-of-the-art (SOTA) results on three of the English datasets. Furthermore, we assessed our approach on three Chinese datasets, and achieved an average improvement of 1.3% over the baseline model. Additionally, in the Few-shot learning experiment, our model outperformed the baseline performance by 5%, especially when the data volume was reduced by around 0.4. Our ablation experiments further validated the effectiveness of STMAP.2

Word segmentation task is the cornerstone of text processing. There are 7111 languages worldwide, most of which are low-resource languages. This paper attempts to solve the problem of multilingual unsupervised word segmentation using common points between languages without tagged corpus. We find that words are only a relationship between phrases and non-phrases in each language, and the frequency of their occurrence obeys the normal distribution. Based on the objective law of language and pre-training language model, this paper defines the concept of Word Influence and designs its calculation formula, and loss function. Combined with the fine-tuning word segmentation task, a multilingual unsupervised word segmentation model was proposed. In order to apply to multiple languages, the model’s key parameters can be learned independently. Its validity and advancement have been proved on Chinese, Japanese, and English data sets. Finally, we discuss the challenges of word segmentation in the pre-trained language model environment.

Many Chinese NER models only focus on lexical and radical information, ignoring the fact that there are also certain rules for the pronunciation of Chinese entities. In this paper, we propose VisPhone, which incorporates Chinese characters’ Phonetic features into Transformer Encoder along with the Lattice and Visual features. We present the common rules for the pronunciation of Chinese entities and explore the most appropriate method to encode it. VisPhone uses two identical cross transformer encoders to fuse the visual and phonetic features of the input characters with the text embedding. A selective fusion module is used to get the final features. We conducted experiments on four well-known Chinese NER benchmark datasets: OntoNotes4.0, MSRA, Resume, and Weibo, with F1 scores of 82.63%, 96.07%, 96.26%, 70.79% respectively, improving the performance by 0.79%, 0.32%, 0.39%, and 3.47%. Our ablation experiments have also demonstrated the effectiveness of VisPhone.

It is a recurrent finding that software development is often troubled by considerable delays as well as budget overruns and several solutions have been proposed in answer to this observation, software fault prediction being a prime example. Drawing upon machine learning techniques, software fault prediction tries to identify upfront software modules that are most likely to contain faults, thereby streamlining testing efforts and improving overall software quality. When deploying fault prediction models in a production environment, both prediction performance and model comprehensibility are typically taken into consideration, although the latter is commonly overlooked in the academic literature. Many classification methods have been suggested to conduct fault prediction; yet associative classification methods remain uninvestigated in this context. This paper proposes an associative classification (AC)-based fault prediction method, building upon the CBA2 algorithm. In an empirical comparison on 12 real-world datasets, the AC-based classifier is shown to achieve a predictive performance competitive to those of models induced by five other tree/rule-based classification techniques. In addition, our findings also highlight the comprehensibility of the AC-based models, while achieving similar prediction performance. Furthermore, the possibilities of cross project prediction are investigated, strengthening earlier findings on the feasibility of such approach when insufficient data on the target project is available.