Combined Image Retrieval (CIR) involves retrieving an image based on a reference image and a brief text description, which is widely present in various scenarios such as fashion recommendation. Existing methods can be mainly divided into two categories, respectively supervised CIR methods and Zero-Shot CIR (ZS-CIR) methods. In contrast to supervised CIR methods, which need manually annotated triples for training task-specific models, ZS-CIR models can be trained using images datasets only and performs well. However, ZS-CIR still faces the primary challenge of learning how to map pseudo-words to images within the joint image-text embedding space. Therefore, in this paper, we propose a novel image-text mapping network, named MLLM-I2W, which adaptively converts description-related image information into pseudo-word markers for precise ZS-CIR. Specifically, the image and text encoding enhancement module within the MLLM prompt selects subject headings and generates text descriptions. It then reduces the modality gap between images and text using uncertainty modeling. An adaptive weighting module and a prototype are proposed to adjust and learn the deep fusion features, which are further mapped to pseudo-word markers via well-designed MOE-based mapping network. Our model demonstrates consistent improvements across common CIR benchmarks, including COCO, CIRR, and Fashion-IQ.

Automatic survey generation has emerged as a key task in scientific document processing.While large language models (LLMs) have shown promise in generating survey texts, the lack of standardized evaluation datasets critically hampers rigorous assessment of their performance against human-written surveys.In this work, we present SurveyGen, a large-scale dataset comprising over 4,200 human-written surveys across diverse scientific domains, along with 242,143 cited references and extensive quality-related metadata for both the surveys and the cited papers.Leveraging this resource, we build QUAL-SG, a novel quality-aware framework for survey generation that enhances the standard Retrieval-Augmented Generation (RAG) pipeline by incorporating quality-aware indicators into literature retrieval to assess and select higher-quality source papers.Using this dataset and framework, we systematically evaluate state-of-the-art LLMs under varying levels of human involvement-from fully automatic generation to human-guided writing.Experimental results and human evaluations show that while semi-automatic pipelines can achieve partially competitive outcomes, fully automatic survey generation still suffers from low citation quality and limited critical analysis 1 .

Terahertz (THz) communications are envisioned as a key scenario for the sixth-generation (6G) wireless communication system, which however suffer from some challenges, such as high path loss, poor power amplifier (PA) efficiency and high phase noise (PN). Low peak-to-average power ratio (PAPR) transmissions can alleviate the poor PA efficiency and improve the cell coverage. To support the low PAPR transmissions, pi/2 BPSK was introduced in the 3GPP 5G NR specifications and has now become mandatory. In addition to that, frequency domain spectrum shaping (FDSS) technique can be applied to the pi/2 BPSK modulation symbols to further reduce the PAPR, while occupies more spectrum resource. In this paper, we proposed a novel waveform based on enhanced discrete Fourier transform spread orthogonal frequency division multiplexing (E DFT-s-OFDM) waveform, in which the FDSS is replaced by the 1+D filtering process in the time domain. The 1+D filter has the characteristic of good time-frequency localization, so lower out-of-band leakage and less interference of the data part of the symbol to the tail sequence are achieved in E DFT-s-OFDM waveform. Compared with the DFT-s-OFDM waveform, the simulation results show that the novel waveform with 1+D filter scheme has lower PAPR, lower out-of-band leakage, and better phase noise estimation and compensation effect.

In this paper, a new data modulation and demodulation scheme are proposed for the terahertz (THz) communications. The new data modulation scheme is named as M-ary pi/2 amplitude phase shift keying (APSK) modulation, which modulates data bits to data symbols with two kinds of phases separated by pi (180°) and then rotates the data symbols by pi/2 in alternating symbol. So the new data modulation scheme can resist the large phase noise (PN) and reduce the peak-to-average power ratio (PAPR). Accordingly, a new data demodulation scheme is also designed to apply for the receiver based on Log-likelihood ratio (LLR) in a polar coordinate instead of conventional cartesian coordinate. The simulation results show that the proposed data modulation and demodulation scheme can achieve superior PAPR and BLER performance in THz frequency bands.

A novel single-carrier waveform scheme is proposed to overcome the main challenges in terahertz (THz) communications such as relatively large path loss, low efficiency of power amplifier and high phase noise. The novel waveform adopts a novel symbol structure with the combined methods of head and tail sequences and frequency domain spectrum shaping (FDSS) together. The head and tail sequences are copied respectively from the head and tail parts of the reference signal symbol. Besides the function of resisting the multipath delay without the use of a conventional cyclic prefix (CP), the sequence can be used to estimate and compensate for the phase noise. By adding FDSS operation, lower peak-to-average power ratio (P APR) and less interference from the data part of the symbol to the tail sequence are achieved. In this paper, the relevant slot structure, frame structure and link block structure of transceiver are also designed. Compared with the 5G waveform DFT-s-OFDM, the proposed waveform scheme improves the spectrum efficiency. The simulation results show that the novel waveform scheme has lower P APR, better phase noise estimation and compensation effect, and lower out-of-band leakage.

In today's dynamic media landscape, products are reviewed by consumers in social media and reported by journalists in traditional media. This paper will focus on the relationship among the two types of “earned” media and product sales. Previous studies have focused on either traditional or social earned media, but rarely both. We will aim to bridge that gap using the following points of analysis: the New York Times Best Seller List as traditional media; Amazon user reviews as social media; and book purchases through Amazon as product sales. We find that: (1) both traditional and social earned media influence sales; (2) sales have a reciprocal effect on social earned media; and (3) traditional and social earned media influence each other. Communication through multiple media is known to produce the “synergy effect” in which one media activity enhances the effect of another. Our results suggest a new benefit unique to the use of multiple earned media. We call this the “multiplier effect,” which occurs when one earned media activity increases the level of another by becoming a “sounding board” that amplifies positive messages, as well as a bridge that allows messages to propagate freely in an interactive media system. Therefore, multiple earned media produce combined sales effects greater than those resulting from the sum of their parts. This analysis supports Amazon's decision to use multiple earned media to benefit from an ecosystem where product sales and earned media both influence and are influenced by one another. The paper will address the implications for marketing communication and media industry.

It has become increasingly important for companies to utilize electronic word of mouth (eWOM) in their marketing campaigns for desired product sales. Identifying key eWOM disseminators among consumers is a challenge for companies. WOM is an interpersonal communication in which a sender spreads a message to receivers. Previously, researchers and practitioners have searched for opinion leaders by examining senders and receivers due to limited records on WOM message. Our study identifies three types of opinion leaders through eWOM using a message-based approach that elicits more accurate and comprehensive information on opinion leadership than sender-based and receiver-based approaches. We demonstrate that eWOM of opinion leaders drives product sales due to their product experience and knowledge background. Our findings suggest that companies can increase product sales via effective use of eWOM of such opinion leaders. Managerial and marketing implications are addressed.