3f:null 40:["$","title",null,{"children":"Paper: Nested cross-validation for sentiment analysis of online reviews: A case study in sustainable tourism | anon author | Observable"}] 46:T32d6,const gridDeck = (() => { const keys = ["sqrt", "log2"]; const colors = {sqrt: "#6f7f91", log2: "#b59b78"}; const textStyle = {fontSize: "14px", color: "#555555"}; // long format: one row per (fold, hyperparameter) const long = grid.flatMap(d => keys.map(k => ({fold: d.Fold, setting: k, value: d[k]}))); // ---------- Slide 1: title only ---------- const slide1 = html`

Nested cross-validation for sentiment analysis of online reviews: A case study in sustainable tourism

Interactive visualisation of figures presented in the results section.

Hyperparameter comparison: accuracy per fold for max_features = sqrt and log2.
Performance metrics: boxplots of accuracy, sensitivity, specificity, HM-SS (harmonic mean of sensitivity and specificity) and AUC-ROC.
Feature importance: the most or least important features of the random forest.

Use the arrow keys or the buttons below to move between slides. Hover over any chart to read exact values, and use the controls above each chart to change what is shown.

`; // ---------- Slide 2: line chart with controls ---------- const seriesInput = Inputs.checkbox(keys, {label: "Settings", value: keys}); const zoomInput = Inputs.toggle({label: "Zoom y axis", value: true}); const lineBox = html`

`; const drawLine = () => { const chosen = seriesInput.value; const subset = long.filter(d => chosen.includes(d.setting)); lineBox.replaceChildren(Plot.plot({ width: 900, height: 400, marginLeft: 60, marginBottom: 50, grid: true, style: {...textStyle, background: "white"}, x: {label: "Fold", labelArrow: "none", ticks: 4, tickFormat: d => String(d)}, y: { label: "Accuracy", labelArrow: "none", domain: zoomInput.value ? [0.92, 0.94] : [0, 1] }, color: {legend: true, domain: keys, range: keys.map(k => colors[k])}, marks: [ Plot.lineY(subset, {x: "fold", y: "value", stroke: "setting", strokeWidth: 2.5}), Plot.dot(subset, {x: "fold", y: "value", stroke: "setting", fill: "white", r: 4}), Plot.tip(subset, Plot.pointerX({ x: "fold", y: "value", stroke: "setting", fontSize: 12, title: d => `${d.setting}\nFold: ${d.fold}\nAccuracy: ${d.value.toFixed(4)}` })) ] })); }; seriesInput.addEventListener("input", drawLine); zoomInput.addEventListener("input", drawLine); drawLine(); const slide2 = html`

Accuracy per fold for each setting

Each point is the accuracy on one fold. Tick or untick a setting to show or hide it. Hover to read exact values. The zoom switch changes the y axis between 0.92-0.94 and the full 0-1 range.

${seriesInput}${zoomInput}

${lineBox}

`; // ---------- Slide 3: boxplots of the nested CV metrics ---------- const metricKeys = ["Accuracy", "Sensitivity", "Specificity", "F-Score", "AUC-ROC"]; const palette = ["#6f7f91", "#b59b78", "#8fa38b", "#a4879a", "#c08a7a"]; // display names: the data column stays "F-Score", only the label shown changes const displayName = {"F-Score": "HM-SS"}; const nameOf = k => displayName[k] ?? k; const rowsRaw = Array.isArray(metrics) ? metrics : metrics.sheet(0, {headers: true}); // the sheet can be the raw runs or a pandas describe() table (count, mean, std, min, 25%, ...) const statLabels = ["count", "mean", "std", "min", "25%", "50%", "75%", "max"]; const labelOf = r => { const v = Object.values(r).find(x => typeof x === "string" && statLabels.includes(x.trim())); return v ? v.trim() : null; }; const isSummary = rowsRaw.some(r => labelOf(r) !== null); let longM = [], statsM, nRuns; if (isSummary) { const byLabel = Object.fromEntries(rowsRaw.filter(r => labelOf(r)).map(r => [labelOf(r), r])); nRuns = byLabel["count"] ? +byLabel["count"][metricKeys[0]] : null; statsM = metricKeys.map(k => ({ metric: k, mean: +byLabel["mean"][k], sd: +byLabel["std"][k], min: +byLabel["min"][k], q1: +byLabel["25%"][k], median: +byLabel["50%"][k], q3: +byLabel["75%"][k], max: +byLabel["max"][k] })); } else { nRuns = rowsRaw.length; longM = rowsRaw.flatMap((r, i) => metricKeys.map(k => ({run: i + 1, metric: k, value: +r[k]}))) .filter(d => Number.isFinite(d.value)); statsM = metricKeys.map(k => { const v = longM.filter(d => d.metric === k).map(d => d.value).sort(d3.ascending); return { metric: k, mean: d3.mean(v), sd: d3.deviation(v) ?? 0, min: v[0], q1: d3.quantile(v, 0.25), median: d3.quantile(v, 0.5), q3: d3.quantile(v, 0.75), max: v.at(-1) }; }); } const metricInput = Inputs.checkbox(metricKeys, {label: "Metrics", value: metricKeys, format: nameOf}); const meanInput = Inputs.toggle({label: "Show mean", value: true}); const boxBox = html`

`; const drawBox = () => { const chosen = metricKeys.filter(k => metricInput.value.includes(k)); const subset = longM.filter(d => chosen.includes(d.metric)); const st = statsM.filter(d => chosen.includes(d.metric)); const lo = d3.min(st, d => d.min) ?? 0; const hi = d3.max(st, d => d.max) ?? 1; const pad = (hi - lo) * 0.08 || 0.01; const yDomain = [lo - pad, hi + pad]; const boxMarks = isSummary ? [ Plot.ruleX(st, {x: "metric", y1: "min", y2: "max", stroke: "#555555"}), Plot.barY(st, {x: "metric", y1: "q1", y2: "q3", fill: "metric", fillOpacity: 0.8, stroke: "#555555"}), Plot.tickY(st, {x: "metric", y: "median", stroke: "#333333", strokeWidth: 2}) ] : [Plot.boxY(subset, {x: "metric", y: "value", fill: "metric", fillOpacity: 0.8, stroke: "#555555"})]; boxBox.replaceChildren(Plot.plot({ width: 900, height: 420, marginLeft: 60, marginBottom: 40, grid: true, style: {...textStyle, background: "white"}, x: {label: null, domain: chosen, padding: 0.4, tickFormat: nameOf}, y: { label: "Value", labelArrow: "none", domain: yDomain }, color: {domain: metricKeys, range: palette}, marks: [ ...boxMarks, meanInput.value ? Plot.dot(st, {x: "metric", y: "mean", symbol: "diamond", fill: "white", stroke: "#333333", r: 5}) : null, Plot.tip(st, Plot.pointerX({ x: "metric", y: "median", maxRadius: 300, fontSize: 12, title: d => `${nameOf(d.metric)}\nMean: ${d.mean.toFixed(4)}\nSD: ${d.sd.toFixed(4)}\nMin: ${d.min.toFixed(4)}\nQ1: ${d.q1.toFixed(4)}\nMedian: ${d.median.toFixed(4)}\nQ3: ${d.q3.toFixed(4)}\nMax: ${d.max.toFixed(4)}` })) ] })); }; metricInput.addEventListener("input", drawBox); meanInput.addEventListener("input", drawBox); drawBox(); const slide3 = html`

Distribution of the performance metrics

Boxplots over ${nRuns} runs. The box spans the first to third quartile and the line is the median. ${isSummary ? "Whiskers go from the minimum to the maximum." : "Whiskers extend to 1.5 times the interquartile range and dots are outliers."} Hover over a box to read its statistics. Tick or untick metrics or show the mean (diamond). HM-SS = 2 · Sensitivity · Specificity / (Sensitivity + Specificity); it is not the F1-score.

${metricInput}${meanInput}

${boxBox}

`; // ---------- Slide 4: feature importance ---------- const rowsF = Array.isArray(feature_imp) ? feature_imp : feature_imp.sheet(0, {headers: true}); const allF = rowsF.map(r => ({feature: String(r.Feature), importance: +r.Importance})) .filter(d => Number.isFinite(d.importance)); const minImp = 0.005; // features with importance at or below this value are removed const featF = allF.filter(d => d.importance > minImp).sort((a, b) => d3.descending(a.importance, b.importance)); const topInput = Inputs.range([1, featF.length], {label: "Number of features", step: 1, value: Math.min(20, featF.length)}); const orderInput = Inputs.radio(["Descending", "Ascending"], {label: "Order", value: "Descending"}); const featBox = html`

`; const featNote = html`

`; const featTitle = html`

`; const drawFeat = () => { const n = topInput.value; const asc = orderInput.value === "Ascending"; const pool = asc ? featF.slice().reverse() : featF; const top = pool.slice(0, n); const domain = top.map(d => d.feature); featTitle.textContent = asc ? "Least important features" : "Most important features"; featNote.textContent = `Showing ${n} of ${featF.length} features with importance above ${minImp} (${allF.length - featF.length} features with importance of ${minImp} or less were removed).`; featBox.replaceChildren(Plot.plot({ width: 900, height: Math.max(160, n * 24 + 70), marginLeft: 110, marginRight: 50, style: {...textStyle, background: "white"}, x: {label: "Importance", labelArrow: "none", grid: true}, y: {label: null, domain}, marks: [ Plot.barX(top, {y: "feature", x: "importance", fill: "#6f7f91"}), n <= 30 ? Plot.text(top, {y: "feature", x: "importance", text: d => d.importance.toFixed(4), textAnchor: "start", dx: 4, fontSize: 11}) : null, Plot.tip(top, Plot.pointerY({ y: "feature", x: "importance", fontSize: 12, title: d => `${d.feature}\nImportance: ${d.importance.toFixed(5)}\nRank: ${featF.indexOf(d) + 1}` })), Plot.ruleX([0]) ] })); }; topInput.addEventListener("input", drawFeat); orderInput.addEventListener("input", drawFeat); drawFeat(); const slide4 = html`

${featTitle}

Random forest feature importance. Descending starts from the most important feature and adds less important ones as the number grows. Ascending starts from the least important feature (above the cut-off) and adds more important ones. Hover over a bar to read the exact value and the overall rank.

${featNote}

${topInput}${orderInput}

${featBox}

`; // ---------- Deck shell ---------- const slides = [slide1, slide2, slide3, slide4]; let index = 0; const root = html`

`; const stage = root.querySelector(".stage"); const count = root.querySelector(".count"); const show = i => { index = Math.max(0, Math.min(slides.length - 1, i)); stage.replaceChildren(slides[index]); count.textContent = `${index + 1} / ${slides.length}`; root.querySelector(".prev").disabled = index === 0; root.querySelector(".next").disabled = index === slides.length - 1; }; root.querySelector(".prev").onclick = () => show(index - 1); root.querySelector(".next").onclick = () => show(index + 1); root.querySelector(".fs").onclick = () => root.requestFullscreen?.(); const onKey = e => { if (["INPUT", "SELECT", "TEXTAREA"].includes(e.target.tagName)) return; if (e.key === "ArrowRight" || e.key === " ") { e.preventDefault(); show(index + 1); } if (e.key === "ArrowLeft") { e.preventDefault(); show(index - 1); } }; document.addEventListener("keydown", onKey); if (typeof invalidation !== "undefined") invalidation.then(() => document.removeEventListener("keydown", onKey)); show(0); return root; })(); display(gridDeck);47:[{"id":"c6fa76683a7e4409ac01d03a8004ae6e5a7b7185f4aa41761becae6d56678a8a92a699310ae567de242400b953ca22649685fadbd9cff26446cb4842f9b77af4","href":"https://static.observableusercontent.com/files/c6fa76683a7e4409ac01d03a8004ae6e5a7b7185f4aa41761becae6d56678a8a92a699310ae567de242400b953ca22649685fadbd9cff26446cb4842f9b77af4","name":"grid_results.csv","type":"text/csv","lastModified":1792234089701,"size":1041},{"id":"60c6162cd2fdb076602a879bef495b2535949363485c97a8a4c7d1c2635c8043c758a66da60bbaa5974773afe3dcd1e10409346eddae9529e003318f4c1826cf","href":"https://static.observableusercontent.com/files/60c6162cd2fdb076602a879bef495b2535949363485c97a8a4c7d1c2635c8043c758a66da60bbaa5974773afe3dcd1e10409346eddae9529e003318f4c1826cf","name":"metrics_rf_turismo.xlsx","type":"application/vnd.openxmlformats-officedocument.spreadsheetml.sheet","lastModified":1792234084355,"size":19983},{"id":"184cbcb7ea6ef81dc21541a6499276eda48a9d0937a1166035380a9c7b6ae2244faf8a4280f679206395faf38d7bcb86acd756a32d3f8b787cadc196e8e83470","href":"https://static.observableusercontent.com/files/184cbcb7ea6ef81dc21541a6499276eda48a9d0937a1166035380a9c7b6ae2244faf8a4280f679206395faf38d7bcb86acd756a32d3f8b787cadc196e8e83470","name":"feature_rf_t.xlsx","type":"application/vnd.openxmlformats-officedocument.spreadsheetml.sheet","lastModified":1792234095525,"size":1852517}] 44:{"id":"8e5422796258563c","slug":"ncv-interactive-vis","title":"Paper: Nested cross-validation for sentiment analysis of online reviews: A case study in sustainable tourism","thumbnail":"4ddcdfb6de77bd8573ef08fcf3308615cbb60746b281064bec8ad17dc32682b6","access_level":"public","create_time":"$D2026-10-17T10:41:32.979Z","update_time":"$D2026-10-17T15:24:05.431Z","view_time":null,"first_public_time":"$D2026-10-17T12:55:33.650Z","has_unpublished_edits":null,"fork":false,"stars":0,"starred":false,"owner":{"type":"team","id":"3978fa3cff65daec","login":"anon-reviewer","name":"anon author","avatar_url":"https://avatars-next.observableusercontent.com/9cda5a7414b2af7f0be8ea1ea53fb9db6e586c2b1fb2d7cb8f4fef96d014554e","tier":"free_2026","bio":"","following":false},"version":60,"publish_version":60,"role":null,"authored":false,"authors":[{"type":"individual","id":"9d915d1e2421bbb2","login":"anon-reviewer","name":"anon author","avatar_url":"https://avatars-next.observableusercontent.com/9cda5a7414b2af7f0be8ea1ea53fb9db6e586c2b1fb2d7cb8f4fef96d014554e","tier":null,"bio":"","following":false}],"members":[{"type":"team","id":"3978fa3cff65daec","login":"anon-reviewer","name":"anon author","avatar_url":"https://avatars-next.observableusercontent.com/9cda5a7414b2af7f0be8ea1ea53fb9db6e586c2b1fb2d7cb8f4fef96d014554e","tier":"free_2026","role":"editor"}],"comments":[],"custom_thumbnail":true,"trashed":false,"lapsed":null,"subscribed":null,"forks":0,"collection":null,"type":"notebook","body":{"titling":"auto","title":"Paper: Nested cross-validation for sentiment analysis of online reviews: A case study in sustainable tourism","cells":[{"id":1,"value":"# Paper: Nested cross-validation for sentiment analysis of online reviews: A case study in sustainable tourism\nInteractive visualisation of figures presented in the results section.","pinned":false,"hidden":false,"mode":"md","author":"$undefined","summary":"$undefined","tool":"$undefined","providerMetadata":"$undefined","attachments":"$undefined","output":"$undefined","database":"$undefined"},{"id":2,"value":"// data used for the visualisations\nconst grid = FileAttachment(\"grid_results.csv\").csv({typed: true})\nconst metrics = FileAttachment(\"metrics_rf_turismo.xlsx\").xlsx({typed: true})\nconst feature_imp = FileAttachment(\"feature_rf_t.xlsx\").xlsx({typed: true})","pinned":true,"hidden":false,"mode":"ts","author":"$undefined","summary":"$undefined","tool":"$undefined","providerMetadata":"$undefined","attachments":"$undefined","output":"$undefined","database":"$undefined"},{"id":4,"value":"$46","pinned":true,"hidden":false,"mode":"ts","author":"$undefined","summary":"$undefined","tool":"$undefined","providerMetadata":"$undefined","attachments":"$undefined","output":"$undefined","database":"$undefined"}],"files":"$W47","stdlib":"2"}} 45:"$44:body"

Nested cross-validation for sentiment analysis of online reviews: A case study in sustainable tourism

Interactive visualisation of figures presented in the results section.

Contents

Accuracy per fold for each setting

Distribution of the performance metrics