知我论文雷达 · 隐私与数据说明
版本 1.0.5,2026-10-21。新功能以实际安装的版本为准。
独立模式把论文、检索缓存、收藏、作者图与刷新断点保存在当前浏览器。使用方式和网页标注开关也保存在当前浏览器。没有账户、开发者服务器、遥测或广告。
学术网页标注
开启时,在 Google Scholar 搜索结果、Scholar 作者主页的论文列表和 arXiv 论文详情中读取论文标题、显示的作者、年份及论文标识/链接,用已有论文库和随包基础索引匹配,并显示作者关系与证据。对你正在浏览的页面,不读取登录信息、Cookie 或论文全文,浏览本身也不发起论文源站抓取。只有当你打开某篇已收录论文的证据卡、而它的作者机构还没核验时,扩展才读取一次这篇论文的 arXiv 网页版,只从中提取作者与所属机构,结果保存到当前数据源;同一篇 15 分钟内不重复读取。可随时在“数据与标注”中关闭,页面中的标记会移除。
支持 Google Scholar 的 google.com、google.com.hk、google.com.sg、google.co.uk、google.de、google.fr、google.co.jp、google.cn 搜索页和作者主页的论文列表(/citations?user=…),以及 arxiv.org 的论文详情页。标注只作用于这些页面的顶层文档,不在任意网站运行。
扩展后台查询当前浏览器缓存,或你已选择连接的本机原库。网页只接收当前结果所需的书目信息、作者关系与收藏状态,不会收到整份数据库。点击收藏后才写入该数据源的收藏。
个人候选池与检索词
只有你点击并确认加入后,才保存该条目的可见标题、作者、来源链接,以及你填写的理由和关键词。候选池与批准的个人检索词保存在当前浏览器的扩展存储中,独立于原库或独立论文缓存;手动候选不会自动成为已核验论文或作者证据。
积累到 10 篇未整理记录时,候选池入口及工具栏图标会提示。不会定时抓取、自动调用模型或自动采用建议。本地关键词整理由你手动启动。导出给其他工具的资料包包含选定书目及你的收录理由,仅下载到本机;导入的工具建议需要逐条确认后才能成为个人检索词。
使用个人检索词仍由你发起检索或联网更新,不自动修改公共专题、机构规则或论文评分。卸载扩展或删除浏览器配置也会清除个人候选池与检索词。
研究方向与个人作者
方向包仅包含 arXiv 分类、专题、子专题、词语、优先级和排除项,保存在当前浏览器。导入配置必须先预览、确认;不允许执行代码、设置 API Key 或修改评分公式。试筛可只用缓存;只有选择联网获取时才发送专题检索请求。首次使用时勾选的专题与 arXiv 类目、你新建的专题与子项(名称、词语与优先级)、机构评分偏好以及新手指引的完成进度,同样只保存在当前浏览器;联网更新时,所选专题的词语和类目会作为检索条件发送给 arXiv。机构评分偏好只在本机参与论文排序,不会发送给任何服务。
收录论文时,你可以选择作者并设置个人评级、匹配方式。名单及所选论文依据保存在扩展个人偏好中;不写入共享作者名录。个人加分在工作台综合价值排序中单独使用、显示,可停用。姓名匹配可能存在同名,默认要求共同作者线索;这些偏好不是已核验作者身份。
网页证据卡基于已有书目和作者证据计算原始 Taste 分,标明缺失项;除了上面说的那一次机构核验,不为评分额外抓取数据。导出给 AI 的方向需求文件只下载到本机;插件不替你发送给模型服务。
在 Scholar 作者主页的论文列表上方,扩展显示一个可折叠的“知我作者摘要”:用本页论文标题与你自己的研究方向计算命中比例,并照抄页面上的引用统计与证据覆盖情况。方向数据由工作台在本地发布一份快照(专题名、优先级、词语)到扩展存储,供后台计算使用;网页只收到命中数量、命中的标题,以及你所选方向的名称和各自的词语数量(用来说明比对范围),收不到这份词语表。计算在本机完成,不额外联网,也不读取其他扩展的数据。
打开作者主页时,扩展还会把该作者的姓名、Scholar 主页编号、页面上的引用统计、按署名位置(一作、共同一作、二作、通讯、末位)算出的论文数、引用和 h 指数,以及本页已收录论文的署名位置,保存到当前数据源,用来放缓这些论文的时间折旧;不保存论文列表本身。若页面上有 Author Highlighter 的控件,只检查它是否存在,不读取它的数据。
个人推荐:确认研究方向后,扩展依据已存书目、所选方向与优先级自动计算私人作者权重,并形成每篇论文的个人分。可在“数据与标注”中关闭自动作者加权;基础证据分与已有手动关注分别保留。身份不明或书目不足时保持中性。计算只读取当前数据源与本地积累,不额外访问外网或调用模型,不新增持久论文副本。私人权重与个人分不发送给开发者、官网或学术网页。
研究图谱与本地演化
研究图谱从既有公开书目和当前数据源的作者证据生成可交互的局部视图。共同署名是连线依据,点击作者可查看论文来源。姓名未消歧时保留为单篇署名节点;截断署名不生成新合作关系,大型团队的全连线不展开。节点与连线不是导师关系、全球排名或质量评价。
“启用本地演化”默认关闭。你主动开启后,仅在你实际打开且启用标注的 Scholar 作者主页上,保存公开作者姓名、主页编号、完整可见署名、论文标题、年份及引用条目标识;已收录论文只保存 ID、版本与作者位置引用。不读取其他网站、全局浏览历史、账户或 Cookie,不后台翻页或抓取论文,不发送给开发者及模型服务。
本地关注按作者每天最多增加一次,并按 30 天半衰期淡化,关注次数只影响图中的光晕;积累的公开书目可补充私人作者权重,再根据你的研究方向进入论文个人分。仅保存聚合日期与关注强度,不保存逐次访问日志。记录保存在当前浏览器,独立于原库,最多保留 120 位作者、480 条论文线索、每位作者 40 条线索,总上限 512 KB,较早记录按该边界滚动保留。
可随时暂停,保留现有记录并停止积累;也可确认“清除本地演化”,清除关注与相关新增线索,并暂停采集。论文库、收藏和原有作者统计保留。卸载扩展或删除浏览器配置也会清除本地演化记录。官网图谱仅使用公开书目与预置演示,不读取或记录访问者的插件数据、浏览习惯。
基础索引与更新
安装包附带有限范围的公开论文书目及作者关系证据,不含个人收藏、笔记、检索历史或本机路径。外部作者标识来自公开作者资料。设置中显示数据版本、数量和最新首发日期;基础索引不是实时或完整的 arXiv 数据库。用户的后续检索缓存可增加识别范围。
联网检索由你发起,会将检索词、论文或作者标识发送给 arXiv、OpenAlex 或 Crossref。它们会接收到连接所需的网络信息。机构图标使用随包资源,论文或 PDF 链接由浏览器打开。基础功能无需模型 API Key。
原库与备份
“连接原库”是可选功能,只访问本机 127.0.0.1:38741 的 Paper Radar 服务;启用时请求该地址权限。插件不会将你的数据库分发给其他用户。连接中断时可以展示随包基础索引,并明确提示原库未连接;收藏不会悄悄改存到另一个数据源。
右键检索只处理你主动选择并通过菜单提交的文字。缓存不按天数自动删除。卸载插件或删除浏览器配置将清除独立缓存;收藏请通过应用内导出按钮备份。本机桥接库独立保存。
更新日期:2026-10-21。反馈问题可通过应用商店的开发者联系入口。
作者卡片可展示公开署名机构、历史关联及雷达作者先验分档。分档属于本工具的内部参考,不代表全球排名;资料不足时保留未评级。公共基础索引不含私人收藏、笔记、查询或论文原始评分。
来源与致谢
机构名称与标志归各自所有者,仅用于识别,不代表合作或背书。
Thank you to arXiv for use of its open access interoperability.
Paper Radar · Privacy and data
Policy version 1.0.5 · October 21, 2026. Feature availability depends on the installed release.
Standalone mode stores papers, search cache, saved papers, author graphs and refresh checkpoints in the current browser. Usage settings and annotation switches also stay in the browser. There is no account system, developer server, telemetry or advertising.
Annotations on scholarly pages
When enabled, the extension reads paper titles, displayed authors, years and paper identifiers or links from Google Scholar search results, paper lists on Scholar author profiles and arXiv abstract pages. It matches these against existing records and the limited bundled index to display author relationships and evidence. It does not read login information, cookies or full paper text. Browsing alone does not fetch papers from source sites. Only when you open an evidence card for an already indexed paper whose affiliations have not been checked, the extension reads that paper's arXiv HTML once to extract authors and affiliations, saving the result to the current data source. The same paper is not checked again within 15 minutes. You can disable annotations in Data and annotations, which removes the marks from pages.
Supported Scholar domains are google.com, google.com.hk, google.com.sg, google.co.uk, google.de, google.fr, google.co.jp and google.cn. The extension supports search pages and author paper lists at /citations?user=…, as well as arxiv.org abstract pages. Annotations operate only in the top-level document of these supported pages, not on arbitrary websites.
The background queries the current browser cache or the original local library you chose to connect. A webpage receives only the bibliography, author relationships and saved status needed for its current results, not the complete database. A paper is saved to that data source only after you click Save.
Personal candidates and search terms
Only after you explicitly confirm an addition does the extension store the visible title, authors, source link, your reason and your keywords. Candidates and approved personal search terms stay in extension storage in the current browser, separately from the original library or standalone paper cache. Manually added candidates do not automatically become verified papers or author evidence.
After ten unorganized candidates accumulate, the candidate entry and toolbar icon show a reminder. The extension does not periodically crawl, automatically call a model or automatically adopt suggestions. Local keyword organization is started manually. A package exported for another tool contains the selected bibliography and your reasons, and is only downloaded locally. Imported suggestions require individual confirmation before becoming personal search terms.
Using personal search terms still requires you to start a search or online refresh. It does not change public topics, institution rules or paper scoring automatically. Uninstalling the extension or removing the browser profile also removes personal candidates and search terms.
Research interests and personal authors
Interest packs contain only arXiv categories, topics, subtopics, terms, priorities and exclusions, stored in the current browser. Imports require preview and confirmation; they cannot execute code, configure an API key or modify scoring formulas. Trial filtering can use the cache only. Only an explicitly selected online fetch sends topic queries. Topics and arXiv categories chosen during onboarding, custom topics and subtopics, institution scoring preferences and onboarding progress also remain local. During an online refresh, selected terms and categories are sent to arXiv as query conditions. Institution scoring preferences are used locally for ranking and are not sent to any service.
When adding a paper, you can select authors and set a personal rating and matching method. The list and its paper evidence stay in personal extension preferences, not the shared author catalogue. Personal score additions are shown separately in the workspace and can be disabled. Names can be ambiguous; coauthor clues are required by default. These preferences do not establish a verified author identity.
Evidence cards compute a raw Taste score from existing bibliography and author evidence, marking missing fields. Apart from the single affiliation check described above, scoring does not trigger extra fetching. Direction requirements exported for AI are downloaded locally; the extension does not send them to a model.
Above a Scholar author's paper list, a collapsible KnowMe author summary compares the page titles with your research interests and reproduces the page's citation statistics and evidence coverage. The workspace publishes a local snapshot of topic names, priorities and terms into extension storage for background calculations. The webpage receives only counts, matched titles, the names of your selected interests and their term counts, not the term lists. Computation is local, makes no additional network request and does not read other extensions' data.
Opening an author profile also stores the author's name, Scholar profile identifier, displayed citation statistics, paper counts, citations and h-index by authorship position, and authorship positions on already indexed papers, in the current data source. These records can slow time depreciation for those papers. The paper list itself is not stored. If an Author Highlighter control exists, the extension checks only its presence and does not read its data.
Personal recommendations: after research interests are confirmed, the extension calculates private author weights from stored bibliography, your topics and priorities, and adds their adjustment to each paper’s personal score. Automatic weighting can be disabled in Data and annotations. Base evidence and existing explicit follows remain separate. Ambiguous identities and insufficient bibliography retain neutral weights. Calculation reads existing local evidence only, makes no additional external or model requests, and creates no new persistent paper copy. Private weights and personal scores are not sent to the developer, website or academic webpages.
Research Atlas and local evolution
The interactive map is a selected view of existing public bibliography and author evidence from the current data source. Shared authorship creates links, with source papers available on selection. Unresolved names remain scoped to individual papers. Truncated bylines do not create new collaborations; large-team papers do not expand into full cliques. These links are not supervision relationships, global rankings or quality ratings.
Local evolution is off by default. After you enable it, the extension uses only Scholar author profiles you actually open with annotations enabled. It stores public author names, profile identifiers, complete visible bylines, paper titles, years and citation-entry identifiers. Already indexed papers use compact references to paper IDs, versions and authorship positions. It does not read other websites, browser-wide history, accounts or cookies; it does not crawl pages or papers in the background, or send these observations to the developer or model services.
Attention increases at most once per author per day and fades with a 30-day half-life. Visit counts only affect attention halos. Accumulated public bibliography can extend private author weights, which contribute to personal paper scores according to your research interests. Only aggregate dates and weights are kept, not a log of individual visits. Records stay in this browser, separate from the original library, with limits of 120 profiles, 480 paper clues, 40 clues per profile and 512 KB overall. Older records roll off within those limits.
You can pause collection while retaining existing records. Confirming Clear local evolution removes attention and the associated new clues and pauses collection. It preserves the paper library, saved papers and existing author statistics. Uninstalling the extension or deleting the browser profile removes these local records. The website map uses public bibliography and a preset example; it does not read or record visitors' extension data or browsing habits.
Bundled index and updates
The installation package contains a limited set of public bibliography and author relationship evidence. It contains no personal saved papers, notes, query history or local paths. External author identifiers come from public author information. Settings show the data version, counts and latest first-submission date. The bundled index is neither a real-time nor a complete arXiv database; your subsequent search cache can expand recognition.
Searches you initiate send query terms and paper or author identifiers to arXiv, OpenAlex or Crossref. These services receive the network information needed for the connection. Institution icons are bundled; paper and PDF links open in the browser. Basic functions require no model API key.
Original library and backups
Connecting the original library is optional and accesses only the Paper Radar service at 127.0.0.1:38741, requesting permission for that local address when enabled. Your database is not distributed to other users. If disconnected, the extension can display the bundled index and clearly indicates that the original library is unavailable. It does not silently save papers to another data source.
Context-menu search processes only text you deliberately select and submit through that menu. Cache is not automatically deleted by age. Uninstalling the extension or deleting the browser profile clears the standalone cache. Back up important saved papers with the in-app export control. The local bridge library is stored separately.
Policy updated October 21, 2026. Report issues through the developer contact entry in the Chrome Web Store.
Author cards may display public affiliations, historical associations and internal Radar author-prior tiers. Tiers are references within this tool, not global rankings. Insufficient evidence remains unrated. The public base index contains no private saved papers, notes, queries or original paper scores.
Sources and acknowledgements
Institution names and logos belong to their respective owners. They are used only for identification and do not imply partnership or endorsement.
Thank you to arXiv for use of its open access interoperability.