Group 1: Investment Projects and Trends - In Q1 2025, Guangzhou signed 677 new industrial investment projects with a total expected investment of 227.49 billion yuan, including 422 high-quality projects [1] - The city has seen significant growth in domestic investment, ranking first in total investment and second in growth rate across the province [1] - Major projects in the manufacturing sector, such as a 54.9 billion yuan direct current transmission project and a new automotive technology company, are boosting development confidence [4][5] Group 2: Focus on Emerging Industries - Guangzhou is targeting strategic emerging industries like artificial intelligence and biomanufacturing through "targeted investment," aiming to attract high-tech enterprises [2] - Notable projects include the establishment of the China headquarters for Kasei Rui Biotech and the global headquarters for Micronano Robotics [2] - The digital economy is also expanding, with companies like ByteDance and Alibaba establishing significant operations in the region [2] Group 3: Foreign Investment and Global Engagement - Guangzhou has actively engaged in international investment promotion, hosting multiple delegations to countries like Spain, Japan, and Singapore [6] - The city has been recognized as China's preferred investment destination for eight consecutive years, with a 3.53% year-on-year increase in new foreign investment projects [7] - In Q1, foreign investment in the manufacturing sector reached 4.902 billion yuan, a remarkable increase of 86.8% [7] Group 4: Business Environment and Policy Innovations - Guangzhou has enhanced its business environment by integrating resources and optimizing policies to attract investments [9] - The launch of the "Guangzhou Investment Land Supermarket" allows investors to access real-time information on available land, promoting ease of investment [9] - The city has implemented various measures to streamline project approvals and reduce costs for businesses, addressing common investment challenges [9][10]

Core Insights - ZhenFund ranked first in the "2024 Annual Top 50 Best Early-Stage Venture Capital Institutions in China" by 投中信息 [1][5] - The ranking is based on a comprehensive evaluation of over a thousand active institutions, highlighting ZhenFund's significant influence in the venture capital landscape [4][5] Group 1: Rankings and Recognition - ZhenFund received recognition in four categories, including being the top early-stage venture capital institution [5] - Other notable rankings for ZhenFund include being in the "Top 10 Best Return Early-Stage Venture Capital Institutions" and "Top 30 Best Investment Institutions in Digitalization and Information Industry" [5] - ZhenFund's portfolio companies also achieved recognition, with several listed in the "Top 10 Best Investment Cases" across various sectors [16] Group 2: Fund Management and Investment Cases - ZhenFund manages over ¥12.4 billion, with notable investments in companies like Xiaohongshu and Yixian E-commerce [3] - The second-ranked institution, BlueRun Ventures, manages over ¥15 billion, with investments in companies such as Li Auto and Qingyun Technology [3] - Other institutions in the top rankings include Halong Capital and Ming Shi Venture Capital, managing approximately ¥10 billion and ¥13.7 billion respectively [4]

Core Insights - Xiaohongshu has implemented a robust original content protection initiative, achieving an average of over 6 million actions against content theft per month, with a total of 2.63 million accounts penalized over the past year [1][3] - The platform has introduced an "original certification feature" that has protected 63,000 original authors and penalized 32,000 impersonating accounts and 1.24 million instances of content theft in the last five months [1][3] Group 1 - Xiaohongshu is committed to protecting creators' hard work and encourages the continuous output of high-quality original content [3] - The original protection project has enhanced its ability to identify copied content, optimized recognition standards, and improved algorithmic capabilities while reducing response time to user infringement complaints from several days to within 24 hours [3] - The platform is actively educating users by updating rules and increasing notifications to encourage more creators to shift from copying to original content [3] Group 2 - To further safeguard original rights, Xiaohongshu plans to launch a "note original declaration" feature in May, allowing all users to declare their independently created notes as original, which will then receive a designated "original" label and comprehensive protection [3] - The account original certification will also open a "green channel" for external creators, allowing those with over 10,000 followers on any platform to apply [3] - The head of original protection governance at Xiaohongshu emphasized the value of original authors and the platform's commitment to maintaining a fair and orderly creative ecosystem by rigorously combating copying, plagiarism, and impersonation [3]

4月29日，第八届数字中国建设峰会在福建福州正式开幕。中央网信办网络综合治理局局长张拥军在主论坛上介绍了未成年人网络保护模式的最新情况。 未成年人做了精选，又保证可观的丰富度，都会提供几十万乃至上千万条的内容，涵盖科学素养、历史文化、艺术体育、心理健康等多个方面，且日均以数 万计的条数在增加。 4月29日第八届数字中国建设峰会主论坛，中央网信办网络综合治理局局长张拥军介绍当前网络未保模式的特点。摄影：南都记者李玲 张拥军称，相信随着未来未成年人模式的推广，还会有大量的专属 App 开发上线，提供更多丰富优质的内容。 张拥军提到，去年 11 月，网信办发布了移动互联网未成年人模式检测指南；在各手机厂商、互联网平台的共同努力下，今年 4 月打造了全新的互联网未成 年人模式，其中华为、OPPO 和中兴手机从现在起更新系统即可使用，小米、荣耀和 vivo 在已经发布的新机上可以使用。 中央网信办网络综合治理局局长张拥军。摄影：南都记者李玲 他谈到，当前未成年人网络保护模式有三个显著特点： 首先，操作更加简便，一键进入专属空间。手机开机即呈现显著的未成年人模式提醒，同时还可在手机桌面、系统设置、控制中心等点击进入未成年 ...

Core Viewpoint - Recent updates in apps like Xiaohongshu and Weibo allow users to visualize and adjust their content preferences, addressing concerns about algorithm transparency and user control over recommendations [1][4]. Group 1: User Control Features - Xiaohongshu introduced a "content preference adjustment" feature that displays a pie chart of users' browsing preferences over the past week, allowing them to adjust recommendation intensity across categories such as fashion, food, and travel [2][3]. - Users can choose from three levels of recommendation intensity: "reduce recommendation," "default intensity," and "increase intensity," and can also delete or add content tags [2][3]. - Similar features have been implemented in Weibo and Bilibili, where users can manage their content preferences through a pie chart and adjust recommendation settings [2][3]. Group 2: Algorithm Governance and Concerns - There are growing concerns about algorithmic issues such as "killing familiarity" and "information cocooning," prompting users to call for more control over algorithmic recommendations [4]. - The implementation of the "Internet Information Service Algorithm Recommendation Management Regulations" in 2022 mandates that platforms provide options for users to opt-out of personalized recommendations and manage their user tags [4]. - In response to these concerns, major platforms have introduced options for users to toggle personalized recommendations, aligning with regulatory requirements [5]. Group 3: Regulatory Actions - A joint initiative by four ministries launched in November aims to address issues like "information cocooning" and algorithmic manipulation, with a focus on user rights and algorithm transparency [5]. - The initiative requires platforms to conduct self-assessments and report on their compliance with the regulations, which has led to various governance measures being announced by platforms like Douyin, Weibo, and Xiaohongshu [5].

本文由华东师范大学和小红书联合完成，共同第一作者是华东师范大学在读硕士、小红书 NLP 团队实习生黄文轩和翟子杰，通讯作者是小红书 NLP 团队负责人 曹绍升，以及华东师范大学林绍辉研究员。 多模态大模型（MLLMs）在视觉理解与推理等领域取得了显著成就。然而，随着解码（decoding）阶段不断生成新的 token，推理过程的计算复杂度和 GPU 显存 占用逐渐增加，这导致了多模态大模型推理效率的降低。现有的方法通过减少预填充（prefill）阶段的视觉 token 冗余来实现推理加速。遗憾的是，这种在预填充 阶段实现的视觉 token 稀疏化所带来的加速优势，在解码阶段会逐渐减弱。当解码输出的文本 token 数量增多时，这些方法仍然会遇到性能瓶颈。 为了解决上述问题，团队创新性地提出了一个全新的动态视觉 - 文本上下文稀疏化推理加速框架 ——Dynamic-LLaVA。该框架针对多模态大模型在不同推理模式 下（包括预填充阶段以及有无 KV Cache 的解码阶段），设计了定制化的稀疏化推理方案，以实现多模态大模型的高效推理。实验结果表明，Dynamic-LLaVA 在 几乎不损失视觉理解和生成能力的前提 ...

Core Insights - Xiaohongshu is undergoing organizational restructuring in its e-commerce division, shifting the focus from a standalone buying team to better serve merchants, indicating a strategic pivot in its business model [1][3] - The integration of buying and merchant operations reflects a broader trend in Xiaohongshu's e-commerce strategy, moving towards a merchant-centric approach rather than solely relying on buyers [3][5] - The transition from a buyer-centric model to a lifestyle e-commerce concept suggests an effort to enhance the platform's commercial viability and address challenges faced by the buying model [5][7] Summary by Sections Organizational Changes - Xiaohongshu's e-commerce division is restructuring, with the buying operations now integrated into the merchant operations team, aiming to improve service for merchants [1] - This change does not signify a complete abandonment of the buying model, as it remains a method for merchants to sell products [1] Historical Context - The shift towards integrating buying and merchant operations has been in progress since early 2024, with initiatives during the 618 shopping festival aimed at providing traffic incentives for all merchants [3] - The introduction of the lifestyle e-commerce concept marks a significant evolution in Xiaohongshu's approach, moving beyond the initial focus on buyers [3][5] Market Dynamics - The buying model, while initially promising, faces challenges such as selection errors and false advertising, leading to a cooling attitude towards influencer-driven sales across the e-commerce industry [5] - Xiaohongshu's buying model struggles with scalability, as replicating successful influencers is difficult due to the unique skills and market presence required [5][7] Product Focus - The buying model is more suited for niche, non-standard products that emphasize experience and personal branding, but these products often come with higher supply chain challenges [7] - Xiaohongshu's initial choice to focus on buying was partly due to its weaker supply chain capabilities compared to major competitors like Taobao and JD.com [7] Strategic Implications - The evolution from a buying-centric model to a lifestyle e-commerce framework indicates a strategic effort to broaden participation in the e-commerce ecosystem, involving merchants and operators with stronger operational capabilities [7]

Core Insights - The article discusses the recent trend of algorithm transparency among major Chinese internet platforms, highlighting the shift from "mysterious technology" to a more open approach in algorithm governance [1][2][5] - It emphasizes that while algorithm transparency can enhance user trust and improve platform governance, it also poses challenges to existing business models [1][4][8] Group 1: Algorithm Transparency - Douyin has publicly disclosed its algorithm for the first time, marking a significant move in the industry [2] - Other platforms like Meituan and Xiaohongshu have also initiated similar transparency measures, aiming to give users more control over algorithmic recommendations [2][3] - The trend is seen as a necessary step for internet companies to adapt to regulatory expectations and improve user engagement [1][5] Group 2: Challenges and Opportunities - The shift towards algorithm transparency is viewed as a double-edged sword, potentially disrupting established business models while also fostering a more ethical digital environment [1][3][8] - Concerns about issues like "information cocoon" and "big data discrimination" have prompted regulatory scrutiny, leading to a new governance framework for platform algorithms [5][6] - The article suggests that platforms must balance algorithmic efficiency with ethical considerations to avoid negative societal impacts [7][8] Group 3: Industry Implications - The article notes that algorithm governance will be a critical focus for internet companies by 2025, as they navigate the complexities of user trust and regulatory compliance [1][5] - It highlights the need for platforms to rethink their underlying logic and user expectations in light of algorithmic changes [8] - The consensus among industry experts is that algorithm transparency should not only meet regulatory demands but also enhance the overall user experience and societal values [4][8]

Group 1 - Xiaohongshu's average daily search volume reached 600 million by Q4 2024, with travel-related content accounting for approximately 15% of its traffic [1] - The Chinese homestay market is projected to reach a scale of 42.27 billion yuan in 2024, with financing in the industry hitting a seven-year high of 4.49 billion yuan [1] - The competition for traffic is intensifying as Xiaohongshu's attention surpasses that of Online Travel Agencies (OTAs) [1] Group 2 - Ctrip's annual GMV is 53.3 billion yuan, while Xiaohongshu's is 7.273 billion yuan, indicating a significant difference in scale [2] - Ctrip has 1.65 billion users compared to Xiaohongshu's 350 million users, and Xiaohongshu has a higher monthly active user count of 32 million versus Ctrip's 9 million [2] - Tujia, a key player in the homestay market, is under pressure to leverage Xiaohongshu for traffic as it faces challenges in its business model [7][12] Group 3 - Tujia's peak year was 2019, where it achieved over 10,000 managed properties, but it has since struggled with losses [3][4] - Tujia's operational model has evolved, splitting into an online platform and a property management service, but it has not successfully replicated the success of Airbnb [4][6] - Tujia's recent attempts to engage with Xiaohongshu have been limited and lack urgency until recent competitive pressures emerged [11][12] Group 4 - The competition landscape has shifted, with players like Muniu and Xiaohongshu emerging as significant threats to Tujia [13][20] - The industry faces challenges in achieving sustainable profitability, with cash flow management being critical for survival [14][15] - Muniu has successfully reached profitability and has focused on leveraging its own traffic, while Tujia has struggled to maintain its market position [17][19] Group 5 - The battle for user attention is intensifying, with both Tujia and Muniu vying for the same demographic on Xiaohongshu [22] - The shift from traditional channels to social media platforms like Xiaohongshu reflects changing user behaviors and preferences [23][24] - As competition increases, brands must focus on creating meaningful connections with users rather than relying solely on advertising [25]

E-commerce and New Retail - Alibaba International Station has reached the top of the US App Store shopping chart, reflecting the strong capabilities of Chinese cross-border e-commerce, driven by the "China Factory Revealed" trend on TikTok [1] - Xiaohongshu announced the cancellation of non-compete agreements and the "big-small week" work schedule, aiming to enhance employee relations and attract talent in a competitive internet industry [2] Logistics and Supply Chain - Yunda Co., Ltd. reported a revenue of 48.54 billion yuan and a net profit of 1.91 billion yuan for 2024, with a year-on-year revenue increase of 7.9% and a service volume growth of 26.14% [4] - Zhongtong Express has formed a strategic partnership with New Stone Technology to enhance last-mile delivery and promote digital upgrades in the logistics industry [5][6] - Shentong Express announced a revenue of 47.17 billion yuan and a net profit of 1.04 billion yuan for 2024, with a significant business volume increase of 29.83% [7] - Debon Express reported a revenue of 40.36 billion yuan and a net profit of 0.86 billion yuan for 2024, indicating a focus on express delivery and a competitive edge in the market [8] Life Services - Meituan clarified rumors regarding account bans for using dual apps, emphasizing its commitment to maintaining industry order and protecting its reputation [9] - JD.com announced that its daily order volume for food delivery surpassed 10 million, showcasing rapid growth attributed to promotional strategies [10] - Meituan received the first nationwide low-altitude logistics operation certificate, marking a significant milestone in the development of low-altitude logistics in China [11] Innovation and Investment - Hangzhou Quanzhi Technology completed a multi-million Pre-A round financing, which will be used for core technology development and market expansion in the industrial robotics sector [12]