Search

The research field of Information Retrieval (IR) has evolved significantly, expanding beyond traditional search to meet diverse user information needs. Recently, Large Language Models (LLMs) have demonstrated exceptional capabilities in text understanding, generation, and knowledge inference, opening up exciting avenues for IR research. LLMs not only facilitate generative retrieval but also offer improved solutions for user understanding, model evaluation, and user-system interactions. More importantly, the synergistic relationship among IR models, LLMs, and humans forms a new technical paradigm that is more powerful for information seeking. IR models provide real-time and relevant information, LLMs contribute internal knowledge, and humans play a central role of demanders and evaluators to the reliability of information services. Nevertheless, significant challenges exist, including computational costs, credibility concerns, domain-specific limitations, and ethical considerations. To thoroughly discuss the transformative impact of LLMs on IR research, the Chinese IR community conducted a strategic workshop in April 2023, yielding valuable insights. This paper provides a summary of the workshop’s outcomes, including the rethinking of IR’s core values, the mutual enhancement of LLMs and IR, the proposal of a novel IR technical paradigm, and open challenges.

We proposed Reinforced Dependency Graph for Aspect-based Sentiment Classification (RDGSC), a reinforced dependency graph model for aspect-based sentiment classification. In this framework, we train a policy network using deep reinforcement learning and construct a reinforced dependency graph for aspect-based sentiment classification. The graph attention network is used to fuse the aspect-related information in the text over the reinforced dependency graph. Each contextual representation is given an aspect-related attention weight through a retrieve-based attention mechanism. A refined final representation is obtained for classification and calculating delayed reward to guide the policy network to updates. Extensive experiments were conducted on five publicly available datasets, the results show that our method is superior to all the baseline methods in two evaluation indicators Accuracy and F1.

An economical and effective method is still lacking for cadmium (Cd) toxicity reduction and food product safety improvement in soil–vegetable systems. Therefore, this study aimed to reduce the Cd toxicity to pak choi (Brassica campestris L.) by jointly using passivators and organic fertilizer, highlighting food products’ safety based on pot experiments. The results showed that compared with the control, organic fertilizer decreased the Cd content in edible parts and the soil’s available Cd by 48.4% and 20.9% on average, respectively, due to the 0.15-unit increases in soil pH. Once jointly applied with passivators, the decrements increased by 52.3–72.6% and 32.5–52.6% for the Cd content in edible parts and for the soil’s available Cd, respectively, while the pH increment increased by 0.15–0.46 units. Compared with the control, the transport factor of Cd was reduced by 61.9% and 50.9–55.0% when applying organic fertilizer alone and together with the passivators, respectively. The combination treatment of biochar and organic fertilizer performed the best in decreasing the Cd content in the edible parts and the soil’s available Cd. The combination treatment of fish bone meal and organic fertilizer induced the greatest increases in soil pH. The grey relational analysis results showed that the combination treatment of biochar and organic fertilizer performed the best in reducing the potential Cd pollution risk, thereby highlighting the vegetable food safety. This study provides a potential economical and effective technology for toxicity reduction and food safety in Cd-polluted soil.

The deep convolutional networks have a great success in vision classification tasks. For object detection, tasks are divided into two subtasks: localization and classification. The detectors scan the whole image to generate object proposals relying on the predefined anchors or points, then classify and fine trim the proposals. The localization task plays an important role in object detection. The foreground objects and background can be easily confused under complex scenes in the existing approaches. Thus, the localization task results in a bad influence on the performance of classification problems. In order to enhance the object localization perceived, a novel method called Attentional Feature Perceive and Augmentation (Att-FPA) is constructed, which devotes setting up a feature dual perceive for localization and classification. Firstly, an attention mask branch (with an attention mask) is imported to check out if the detecting area is the background or containing one or more objects. By maintaining the feature representation of background, the attention mask is employed to strengthen the feature representation of foreground objects, which significantly enhances the localization perceived ability of objects. Moreover, Att-FPA supplies a novel way on feature re-extraction and augmentation, which further promotes the performance significantly. In the task of Ms COCO object detection, Att-FPA achieves an outstanding promotion(3.0+ mAP) than baseline (Faster-Rcnn) method. It establishes a new efficient method of feature representation, which outperforms the state-of-the-art models in object detection.

Recent years have witnessed rapid developments on computer vision, however, there are still challenges in detecting tiny objects in a large-scale background. The tiny objects knowledge become sparse and weak due to their tiny size, which makes the tiny objects difficult to be detected with the common approaches. In this paper, a new network named Specific Characteristics based Feature Rescaling and Fusion (SFRF) is designed to detect tiny persons in a broad horizon and massive background. Different from the methods in general, a Nonparametric Adaptive Dense Perceiving Algorithm (NADPA) is designed to automatically select and generate a new resized feature map with the high density distribution of tiny objects. Then, a method called Many-For-One strategy is used for feature fusion of the feature pyramid network (FPN) layers to improve the feature representation and detection. Finally, an ensemble model method named hierarchical Coarse-to-fine mechanism is designed based on the proposed methods to further improve the performance. The experiments demonstrate that the proposed approach achieves an obvious performance improvement on tiny object detection than the existing approaches, and our approach has been awarded as the 1st-place in the first large-scale Tiny Object Detection (TOD) challenge.

Stochastic gradient descent(SGD) is the fundamental sequential method in training large scale machine learning models. To accelerate the training process, researchers proposed to use the asynchronous stochastic gradient descent (A-SGD) method in model learning. However, due to the stale information when updating parameters, A-SGD converges more slowly than SGD in the same iteration number. Moreover, A-SGD often converges to a high loss value and results in lower model accuracy. In this paper, we propose a novel algorithm called Trend-Smooth which can be adapted to the asynchronous parallel environment to overcome the above problems. Specifically, Trend-Smooth makes use of the parameter trend during the training process to shrink the learning rate of some dimensions where the gradients' directions are opposite to the trends of parameters. Experiments on MNIST and CIFAR-10 datasets confirm that Trend-Smooth can accelerate the convergence speed in asynchronous training process. The test accuracy that Trend-Smooth achieves is shown to be higher than other asynchronous parallel baseline methods, and is very close to the SGD method. Moreover, Trend-Smooth can also be combined with other adaptive learning rate methods(like Momentum, RMSProp and Adam) in the asynchronous parallel environment to promote their performance.