Measure the words that define a search result A page title has very little space to explain what a URL offers. The META Title TF-IDF Analyzer focuses on that compact field and shows how LuceneUtil scores its lexical vocabulary. Enter a target page and a search phrase that represents the page’s real job. Yellow marks original search-phrase words, while weighted orange marks Similar Words loaded from the same catalog system used during task calculation. Common stop words remain visible with gray dotted styling but add no credit. This separation lets an editor see the evidence without turning a useful title into a bag of keywords. The analyzer treats straightforward singular and plural forms as the same concept, then applies each Similar Word’s stored relationship weight to its TF-IDF contribution. A closely related term can therefore contribute without receiving the same influence as an original search-phrase word. This remains a lexical calculation: the word must occur in the title. Embedding tools handle broader inferred meaning and paraphrases that do not contain any configured term. Each row identifies its source and weight so the content team can trace the calculation instead of guessing which vocabulary counted.
Read term frequency without chasing repetition Each search-word row reports how many times that word occurs in the title. The term-frequency value uses a square-root curve. The first accurate occurrence matters most, while a second copy supplies much less additional value. This is important because titles are especially vulnerable to awkward duplication. A phrase such as “Nursing Program | Nursing Degree | Nursing School” may repeat a high-value word, but it usually communicates less clearly than a concise title that names the credential, subject, and institution once. Use the occurrence count to locate a genuine omission, not to establish a target density. If a meaningful query word is absent, ask whether adding it would make the title more recognizable and accurate. If the title already names the topic, preserve natural language. The hover on each highlighted word identifies it as counted evidence, and the corresponding term row shows the diminishing-return frequency value. Together they reveal why copying the same word again is rarely the best edit.
Understand why rare words can matter more Inverse document frequency describes how uncommon a search word is among the pages being compared. A term found on every displayed page carries less distinguishing information than a term found on only one. The analyzer shows the number of preview documents containing each word and the resulting rarity value. In target-only mode, the calculation uses one document so an editor can still test changes. In comparison mode, both displayed titles supply the preview context and the rarity values update for the pair. A Ranking Sensor uses its complete modeled corpus, so its production rarity values can differ from a two-page preview. That does not make the preview arbitrary. It makes the edit calculation transparent: both sides use the same visible pages and every input is shown. Use the tool to understand direction and term behavior, then recalculate the Ranking Sensor after publishing to confirm the full-corpus factor score. Avoid comparing raw rarity numbers across unrelated sensors because their page collections represent different search environments.
Read all seven title TF-IDF calculation stages The first row reports search-concept coverage and clause coordination. Coverage is a diagnostic count of how many configured, non-stop-word concepts appear; it helps an editor spot omissions. It is not a hidden multiplier. Market Brew deliberately fixes Lucene clause coordination at 1.0, so a missing concept removes that term’s own evidence but does not apply a second coverage penalty to the entire title score. The second row is the weighted term-evidence total. For every original search word or Similar Word, the analyzer multiplies the square root of its occurrence count by inverse document frequency squared and by its relationship weight. Original search words use full relationship weight. A Similar Word uses its displayed fractional relationship. The term cards expose occurrence count, transformed term frequency, document frequency, IDF, relationship, and resulting contribution before field-wide normalization. The third row is query normalization. The interface shows the weighted query-vector magnitude for transparency, but LuceneUtil fixes the actual query-normalization multiplier at 1.0. It therefore does not raise or lower this title score. The fourth row is title-length normalization: 1 divided by the square root of the number of scored, non-stop words. This limits the advantage of adding generic words merely to create more matching opportunities. It does not declare a perfect title length; necessary brand, location, credential, and product words still belong in an accurate title. The fifth row multiplies the weighted term total by coordination, query normalization, and title-length normalization to produce raw TF-IDF additive points. A standalone run stops at this editable preview because it has no task corpus. The sixth row shows the calibrated page-title field boost supplied by a generated task link. The seventh row shows final task-score normalization: the page’s raw Lucene title contribution divided by the strongest raw title contribution in that task’s modeled result set, multiplied by 100. A task score of 100/100 therefore identifies the corpus leader; it does not certify a perfect title.
Compare titles as editorial evidence Authenticated comparison mode places the target and another indexed URL side by side. The outperformer is evidence, not a writing template. It may contain a useful qualifier that the target omits, but it may also use a different brand, location, credential, or page purpose. Compare the highlighted terms, field lengths, coverage diagnostics, and individual contribution rows. Identify the communication difference that is relevant to your page rather than copying the other title’s construction. A stronger comparison title often states the page type earlier, removes vague promotional language, or connects a subject with a meaningful modifier. Translate that lesson into original wording. Preserve target-specific facts and brand conventions. If both titles cover the same query words, focus on clarity instead of forcing a statistical difference. Other ranking factors, search intent, authority, and user behavior can explain the ranking gap. This analyzer deliberately isolates only page-title TF-IDF.
Test one title change before publishing Edit the highlighted target title directly and click Measure This Change. The score card, term rows, heatmap, word count, and calculation update from the draft without publishing content or requesting a crawl. Predict the expected outcome first. Adding a missing meaningful word should change its term frequency, contribution, and coverage diagnostic. Removing an unscored stop word may change presentation but not the scored word count. Repeating an existing term should show diminished benefit. A predictable result confirms that the tool and the proposed edit are measuring the same thing. After measuring, inspect the title as a user would see it. Confirm that it identifies the page, avoids unsupported claims, remains distinct from other pages, and is not overloaded with separators. Consider likely search-result truncation, but do not sacrifice meaning to hit one universal character count. Search interfaces can rewrite or display titles differently. The page’s stored title should still stand on its own as an accurate label in a browser tab, bookmark, share preview, accessibility context, and search result.
Use TF-IDF with the rest of the content toolkit TF-IDF is a lexical diagnostic, not a complete quality score. Pair it with Query Coverage when you need a simpler presence check, Query Proximity when word order and closeness matter, and semantic visualizers when related meaning should count without exact wording. Use the BM25 tools when you want a different frequency model with explicit term saturation and document-length behavior. These algorithms may agree that a title is weak for different reasons, or they may disagree because each isolates a different signal. Keep a short record of the calibrated query, original title, edited title, preview score, and editorial reason. After publication and recrawl, compare the stored title with the approved draft and inspect the recalibrated Ranking Sensor task. A useful workflow connects the task score to visible term evidence, a small defensible edit, and a full-corpus verification. The goal is not a perfect meter. It is a title that tells both people and the modeled search engine exactly what the page is about.
