Zhihao Fan

Zhihao Fan

Auto-generated profile, based on publicly available information

Zhihao Fan

School of Data Science Fudan University

Research Focus & Metrics

Academic Career Length

7Years (2020-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
19 October 2023 in IEEE/ACM Transactions on Audio, Speech, and Language Processing
Verified
Journal Article
Closed access
SWSiyuan Wang
Siyuan Wang
ZWZhongyu Wei
Zhongyu Wei
JXJiarong Xu
Jiarong Xu
TLTaishan Li
Taishan Li
ZFZhihao Fan
Zhihao Fan
Recent pre-trained language models (PLMs) equipped with foundation reasoning skills have shown remarkable performance on downstream complex tasks. However, the significant structure reasoning skill has been rarely studied, which involves modeling implicit structure information within the text and performing explicit logical reasoning over them to deduce the conclusion. This paper proposes a unified learning framework that combines explicit structure reasoning and language pre-training to endow PLMs with the structure reasoning skill. It first identifies several elementary structures within contexts to construct structured queries and performs step-by-step reasoning along the queries to identify the answer entity. The fusion of textual semantics and structure reasoning is achieved by using contextual representations learned by PLMs to initialize the representation space of structures, and performing stepwise reasoning on this semantic representation space. Experimental results on four datasets demonstrate that the proposed model achieves significant improvements in complex reasoning tasks involving diverse structures, and shows transferability to downstream tasks with limited training data and effectiveness for complex reasoning of KGs modality.

1 January 2020
Verified
Conference Paper
Open access
PDF file
ZFZhihao Fan
Zhihao Fan
YGYeyun Gong
Yeyun Gong
ZWZhongyu Wei
Zhongyu Wei
SWSiyuan Wang
Siyuan Wang
YHYameng Huang
Yameng Huang
RZRuofei Zhang
Ruofei Zhang
Zhihao Fan, Yeyun Gong, Zhongyu Wei, Siyuan Wang, Yameng Huang, Jian Jiao, Xuanjing Huang, Nan Duan, Ruofei Zhang. Proceedings of the 28th International Conference on Computational Linguistics. 2020.

28 October 2024
Verified
Conference Paper
Open access
ZLZejun Li
Zejun Li
YWYe Wang
Ye Wang
MDMengfei Du
Mengfei Du
QLQingwen Liu
Qingwen Liu
BWBinhao Wu
Binhao Wu
XHXuanjing Huang
Xuanjing Huang
Recent years have witnessed remarkable progress in the development of large vision-language models (LVLMs). Benefiting from the strong language backbones and efficient cross-modal alignment strategies, LVLMs exhibit surprising capabilities to perceive visual signals and perform visually grounded reasoning. However, the capabilities of LVLMs have not been comprehensively and quantitatively evaluated. Most existing multi-modal benchmarks require task-oriented input-output formats, posing great challenges to automatically assess the free-form text output of LVLMs. To effectively leverage the annotations available and reduce the manual efforts required for constructing new benchmarks, we propose to re-formulate existing benchmarks into unified LVLM-compatible formats. Through systematic data collection and reformulation, we present ReForm-Eval benchmark, offering substantial data for evaluating various capabilities of LVLMs. Through extensive experiments and analysis in ReForm-Eval, we demonstrate the comprehensiveness and reliability of ReForm-Eval in assessing various LVLMs. Our benchmark and evaluation framework is now available at https://github.com/FudanDISC/ReForm-Eval

1 January 2025 in Data Intelligence
Verified
Journal Article
Open access
PDF file
QLQingwen Liu
Qingwen Liu
ZLZejun Li
Zejun Li
ZFZhihao Fan
Zhihao Fan
CYCunxiang Yin
Cunxiang Yin
YHYancheng He
Yancheng He
ZWZhongyu Wei
Zhongyu Wei
Event extraction extracts event frames from text, while grounded situation recognition detectsevents in images. As real-world applications frequently encounter a multitude of unforeseenevents, certain researchers have introduced cross-domain and in-domain event extraction, whilegrounded situation recognition primarily explores in-domain scenarios. Therefore, in this paper,we propose cross-domain grounded situation recognition and establish a new benchmark SWiGXD. In this more challenging setting, we deepen the connection between the two tasks based ontheir underlying unity in two different modalities and explore how to transfer the generalizationability from text to images. Firstly, we utilize ChatGPT to automatically generate textual data,which can be divided into two categories. One category is directly matched with images, establishing a direct connection with the images. The other category encompasses all event types andpossesses greater generalization. Then we employ a uniffed model framework to establish theassociation between textual concepts and local image features and achieve cross-domain generalization transfer across modalities through modality-shared prompts and self-attention mechanism. Furthermore, we incorporate textual data with higher generalization to further assist inimproving generalization on images. The experimental results on the newly constructed benchmark demonstrate the effectiveness of our method.

4 August 2022 in BCP Business & Management
Verified
Journal Article
Open access
PDF file
WFWeipei Fu
Weipei Fu
QLQiannuo Li
Qiannuo Li
MLMengxue Li
Mengxue Li
ZWZhongyu Wang
Zhongyu Wang
ZFZhihao Fan
Zhihao Fan
As the necessities of production and life, construction and plate have always played an important role in China's national economy. It is also an important industry affecting China's economic development. The unit price of raw materials and the cost of transportation and storage have always been the main expenses of enterprises. Therefore, it is particularly important to study how to reduce the ordering cost and transshipment cost of raw materials. This paper makes a quantitative analysis on the supply characteristics of 402 suppliers. According to the supply quantity of suppliers, the orders with an order quantity of 0 are excluded, so as to extract the variance, range and real average value of each supplier's orders in five years. The difference between the enterprise's order quantity and the supplier's supply quantity is extracted, so that they represent four indicators: supplier reliability, sustainability and stability, And the supply intensity of the supplier. After the data is normalized through AHP, the weight of the index is assigned, the index is weighted in combination with TOPSIS, and finally the score is calculated. Then, the 0-1 programming model is established. The coefficient of the objective function is the maximum supply of each supplier every week. After analysis, it is concluded that the maximum supply is two-thirds of the average value plus one-third of the maximum value. The constraint conditions are established so that the sum of the supply of the selected suppliers is greater than the weekly capacity required by the enterprise. The 240 weeks are divided into five years. Considering the influence of practical factors, when predicting the supply of suppliers in the next 24 weeks, only the first half of each year is selected, and the grey prediction method is used to predict the average weekly supply of each supplier in the next 24 weeks. After the prediction, the grey prediction of the existing data is not ideal, and the BP neural network is used to re predict this kind of data. After the prediction, a multi-objective programming model is established to ensure the minimum sum of ordering cost and transshipment cost. Under this scheme, the minimum supply corresponding to a, B and C is calculated.

1 November 2024
Verified
Book Chapter
Closed access
PDF file
ZFZhihao Fan
Zhihao Fan
ZLZejun Li
Zejun Li
SWSiyuan Wang
Siyuan Wang
ZWZhongyu Wei
Zhongyu Wei
HSHaijun Shan
Haijun Shan