diff --git a/packages/frontend/src/workers/cluster.worker.js b/packages/frontend/src/workers/cluster.worker.js
index eff186a8..e5c24776 100644
--- a/packages/frontend/src/workers/cluster.worker.js
+++ b/packages/frontend/src/workers/cluster.worker.js
@@ -1,6 +1,6 @@
 /* eslint no-restricted-globals: 0 */
 import { kruskalMST } from 'visual-insights';
-
+const PearsonThreshold = 0.5;
 function sum (arr) {
   let ans = 0;
   let len = arr.length;
@@ -15,7 +15,7 @@ const cluster = (e) => {
     const { spaces, maxGroupNumber } = e.data;
     let result = [];
     for (let space of spaces) {
-      const { edgesInMST, groups } = kruskalMST(space.matrix, maxGroupNumber);
+      const { edgesInMST, groups } = kruskalMST(space.matrix, maxGroupNumber, PearsonThreshold);
       let measureGroups = new Map();
       for (let i = 0; i < groups.length; i++) {
         if (!measureGroups.has(groups[i])) {
diff --git a/packages/visual-insights/src/dashboard/index.ts b/packages/visual-insights/src/dashboard/index.ts
index 0fa9ecc8..75cd286e 100644
--- a/packages/visual-insights/src/dashboard/index.ts
+++ b/packages/visual-insights/src/dashboard/index.ts
@@ -1,9 +1,10 @@
 import { FieldsFeature, correlation, linearMapPositive } from "../insights/impurity";
 import { DataSource, OperatorType } from "../commonTypes";
-import cluster, { kruskalMST } from "../insights/cluster";
+import cluster from "../insights/cluster";
 import aggregate from 'cube-core';
 import { normalize, entropy } from "../impurityMeasure";
 import { crammersV } from './utils';
+import { CrammersVThreshold, PearsonCorrelation } from '../insights/config';
 
 interface DashBoardSpace {
   dimensions: string[];
@@ -48,7 +49,8 @@ export function getDashBoardSubspace (dataSource: DataSource, dimensions: string
   const measureGroups = cluster({
     matrix: correlationMatrix,
     measures,
-    groupMaxSize: Math.round(measures.length / 6) // todo: make a config: max 6 measures in a dashboard
+    groupMaxSize: Math.round(measures.length / 6), // todo: make a config: max 6 measures in a dashboard
+    threshold: PearsonCorrelation.weak
   })
 
   const dimCorrelationMatrix = dimensions.map(d => dimensions.map(d => 0));
@@ -119,7 +121,8 @@ export function getDashBoardView (dashBoardSpace: DashBoardSpace, dataSource: Da
   const measureGroups = cluster({
     matrix: dashBoardSpace.correlationMatrix,
     measures: measures,
-    groupMaxSize: Math.round(measures.length / 3) // todo: make a config: max 3 measures in a chart
+    groupMaxSize: Math.round(measures.length / 3), // todo: make a config: max 3 measures in a chart
+    threshold: PearsonCorrelation.strong
   });
   for (let group of measureGroups) {
     const meaIndexList = group.map(mea => measures.indexOf(mea))
@@ -145,7 +148,8 @@ export function getDashBoardView (dashBoardSpace: DashBoardSpace, dataSource: Da
     matrix: dimensionCorrelationMatrix,
     measures: dimensions,
     groupMaxSize: 2, // todo: make a config: max 2 dimensions in a chart
-    limitSize: true
+    limitSize: true,
+    threshold: CrammersVThreshold
   })
 
   const dimGroupEntropyMatrix = getEntropyMatrix(dimensionGroups, measures, dataSource);
@@ -173,4 +177,6 @@ function minIndex(arr: number[]) {
     }
   }
   return pos;
-}
\ No newline at end of file
+}
+
+export { crammersV }
\ No newline at end of file
diff --git a/packages/visual-insights/src/dashboard/utils.ts b/packages/visual-insights/src/dashboard/utils.ts
index 7da9dc40..c4aab723 100644
--- a/packages/visual-insights/src/dashboard/utils.ts
+++ b/packages/visual-insights/src/dashboard/utils.ts
@@ -1,45 +1,109 @@
 import { DataSource } from "../commonTypes";
-
-export function chiSquared(matrix: number[][] = [[]]): number {
-  let rowSums = matrix.map(m => 0);
-  let colSums = matrix[0].map(m => 0);
+type nestTree = Map<string, Map<string, number>>;
+export function chiSquared(nestTree: nestTree, xSet: Set<string>, ySet: Set<string>): number {
+  if (typeof nestTree === 'undefined' || typeof xSet === 'undefined' || typeof ySet === 'undefined') {
+    return 0;
+  }
+  let rowSums = new Map<string, number>();
+  let colSums = new Map<string, number>();
   let totalSum = 0;
-  for (let i = 0; i < matrix.length; i++) {
-    for (let j = 0; j < matrix[i].length; j++) {
-      rowSums[i] += matrix[i][j];
-      colSums[j] += matrix[i][j];
-      totalSum += matrix[i][j];
+  for (let x of xSet) {
+    rowSums.set(x, 0);
+  }
+  for (let y of ySet) {
+    colSums.set(y, 0);
+  }
+  for (let [x, node] of nestTree) {
+    for (let [y, counter] of node) {
+      rowSums.set(x, rowSums.get(x) + counter);
+      colSums.set(y, colSums.get(y) + counter);
+      totalSum += counter;
     }
   }
+
   let chis = 0;
-  for (let i = 0; i < matrix.length; i++) {
-    for (let j = 0; j < matrix[i].length; j++) {
-      let observed = matrix[i][j];
-      let expected = rowSums[i] * colSums[j] / totalSum;
+  for (let [x, node] of nestTree) {
+    for (let [y, observed] of node) {
+      let expected = rowSums.get(x) * colSums.get(y) / totalSum;
       chis += (observed - expected) ** 2 / expected;
     }
   }
   return chis;
 }
 
-export function crammersV(dataSource: DataSource, fieldX: string, fieldY: string) {
-  const xSet = new Set(dataSource.map(d => d[fieldX]))
-  const ySet = new Set(dataSource.map(d => d[fieldY]))
-  const xMembers = [...xSet];
-  const yMembers = [...ySet];
-  let xDict = {};
-  let yDict = {};
-  for (let i = 0; i < xMembers.length; i++) {
-    xDict[xMembers[i]] = i;
-  }
-  for (let i = 0; i < yMembers.length; i++) {
-    yDict[yMembers[i]] = i;
-  }
-  let matrix: number[][] = xMembers.map(x => yMembers.map(y => 0));
-  for (let record of dataSource) {
-    matrix[xDict[record[fieldX]]][yDict[record[fieldY]]]++;
+
+export function crammersV(dataSource: DataSource, fieldX: string, fieldY: string): number {
+  const xSet = new Set<string>()
+  const ySet = new Set<string>()
+  const nestTree = new Map<string, Map<string, number>>();
+  let len = dataSource.length;
+  for (let i = 0; i < len; i++) {
+    let record = dataSource[i];
+    xSet.add(record[fieldX])
+    ySet.add(record[fieldY]);
+    if (!nestTree.has(record[fieldX])) {
+      nestTree.set(record[fieldX], new Map());
+    }
+    let node = nestTree.get(record[fieldX]);
+    if (!node.has(record[fieldY])) {
+      node.set(record[fieldY], 0);
+    }
+    node.set(record[fieldY], node.get(record[fieldY]) + 1);
   }
-  const chis = chiSquared(matrix);
-  const V = Math.sqrt(chis / (dataSource.length * Math.min(xMembers.length - 1, yMembers.length - 1)))
+  const chis = chiSquared(nestTree, xSet, ySet);
+  const V = Math.sqrt(chis / (dataSource.length * Math.min(xSet.size - 1, ySet.size - 1)))
   return V;
-}
\ No newline at end of file
+}
+
+
+// can be used for test
+// export function crammersV(dataSource: DataSource, fieldX: string, fieldY: string): number {
+//   const xSet = new Set(dataSource.map(d => d[fieldX]))
+//   const ySet = new Set(dataSource.map(d => d[fieldY]))
+//   const xMembers = [...xSet];
+//   const yMembers = [...ySet];
+//   let xDict = {};
+//   let yDict = {};
+//   for (let i = 0; i < xMembers.length; i++) {
+//     xDict[xMembers[i]] = i;
+//   }
+//   for (let i = 0; i < yMembers.length; i++) {
+//     yDict[yMembers[i]] = i;
+//   }
+//   // let matrix: number[][] = xMembers.map(x => yMembers.map(y => 0));
+//   let matrix: number[][] = [];
+//   for (let  i = 0; i < xMembers.length; i++) {
+//     matrix.push([]);
+//     for (let j = 0; j < yMembers.length; j++) {
+//       matrix[i].push(0);
+//     }
+//   }
+//   for (let record of dataSource) {
+//     matrix[xDict[record[fieldX]]][yDict[record[fieldY]]]++;
+//   }
+//   const chis = chiSquared(matrix);
+//   const V = Math.sqrt(chis / (dataSource.length * Math.min(xMembers.length - 1, yMembers.length - 1)))
+//   return V;
+// }
+
+// export function chiSquared(matrix: number[][] = [[]]): number {
+//   let rowSums = matrix.map(m => 0);
+//   let colSums = matrix[0].map(m => 0);
+//   let totalSum = 0;
+//   for (let i = 0; i < matrix.length; i++) {
+//     for (let j = 0; j < matrix[i].length; j++) {
+//       rowSums[i] += matrix[i][j];
+//       colSums[j] += matrix[i][j];
+//       totalSum += matrix[i][j];
+//     }
+//   }
+//   let chis = 0;
+//   for (let i = 0; i < matrix.length; i++) {
+//     for (let j = 0; j < matrix[i].length; j++) {
+//       let observed = matrix[i][j];
+//       let expected = rowSums[i] * colSums[j] / totalSum;
+//       chis += (observed - expected) ** 2 / expected;
+//     }
+//   }
+//   return chis;
+// }
\ No newline at end of file
diff --git a/packages/visual-insights/src/index.ts b/packages/visual-insights/src/index.ts
index 6286bb77..7e6422c2 100644
--- a/packages/visual-insights/src/index.ts
+++ b/packages/visual-insights/src/index.ts
@@ -7,7 +7,7 @@ import * as Distribution from './distribution';
 
 import * as ImpurityMeasure from './impurityMeasure';
 
-import getInsightViews, { analysisDimensions, getCombination, clusterMeasures, kruskalMST } from './insights/index';
+import getInsightViews, { analysisDimensions, getCombination, clusterMeasures, kruskalMST, getDimSetsBasedOnClusterGroups } from './insights/index';
 import * as Cleaner from './cleaner/index';
 
 import * as UnivariateSummary from './univariateSummary/index'
@@ -29,6 +29,7 @@ export {
   Cleaner,
   getInsightViews,
   getCombination,
+  getDimSetsBasedOnClusterGroups,
   clusterMeasures,
   kruskalMST
 }
\ No newline at end of file
diff --git a/packages/visual-insights/src/insights/cluster.ts b/packages/visual-insights/src/insights/cluster.ts
index cdd2c5e6..705c1ad3 100644
--- a/packages/visual-insights/src/insights/cluster.ts
+++ b/packages/visual-insights/src/insights/cluster.ts
@@ -55,7 +55,7 @@ function unionWithEffect (parents: number[], sizes: number[], n1: number, n2: nu
  * @param matrix adjmatrix
  * @param groupNumber number of group generated by clustering
  */
-function kruskal(matrix: number[][], groupNumber: number): Map<number, number[]> {
+function kruskal(matrix: number[][], groupNumber: number, threshold: number | undefined = 0): Map<number, number[]> {
   const edges = turnAdjMatrix2List(matrix);
   edges.sort((a, b) => b[1] - a[1]);
   const parents = matrix.map((m, i) => i);
@@ -68,7 +68,7 @@ function kruskal(matrix: number[][], groupNumber: number): Map<number, number[]>
       parents[i] = find(parents, i)
     }
     let set = new Set(parents);
-    if (set.size <= groupNumber){
+    if (set.size <= groupNumber || edge[1] < threshold){
       break;
     }
   }
@@ -127,7 +127,6 @@ export function kruskalMSTWithLimitSize(matrix: number[][], limitSize: number =
   const parents = matrix.map((m, i) => i);
   const cloneParents = matrix.map((m, i) => i);
   const sizes = matrix.map(() => 1);
-  let inCutEdge = false;
   for (let edge of edges) {
     if (findWithEffect(parents, sizes, edge[0][0]) !== findWithEffect(parents, sizes, edge[0][1])) {
       if (sizes[edge[0][0]] + sizes[edge[0][1]] > limitSize) {
@@ -159,7 +158,7 @@ export function kruskalMSTWithLimitSize(matrix: number[][], limitSize: number =
  * @param matrix 
  * @param groupNumber number of group generated by clustering
  */
-export function kruskalMST(matrix: number[][], groupNumber: number = 4) {
+export function kruskalMST(matrix: number[][], groupNumber: number = 4, threshold: number | undefined = 0) {
   const edges = turnAdjMatrix2List(matrix);
   edges.sort((a, b) => b[1] - a[1]);
 
@@ -180,7 +179,7 @@ export function kruskalMST(matrix: number[][], groupNumber: number = 4) {
     let set = new Set(parents);
     // TODO:
     // + use kruskalMST instead of kruskal.
-    if (set.size <= groupNumber) {
+    if (set.size <= groupNumber || edge[1] < threshold) {
       inCutEdge = true;
     } else {
       groups = [...parents]
@@ -199,15 +198,16 @@ interface ClusterProps {
   method?: string;
   groupMaxSize?: number;
   limitSize?: boolean;
+  threshold?: number;
 }
 
-function cluster ({ matrix, measures ,method = 'kruskal', groupMaxSize = 4, limitSize = false }: ClusterProps): string[][] {
+function cluster ({ matrix, measures, method = 'kruskal', groupMaxSize = 4, limitSize = false, threshold = 0 }: ClusterProps): string[][] {
   // const groups = kruskal({ matrix, groupMaxSize });
   let groups;
   if (limitSize) {
     groups = kruskalWithLimitSize(matrix, groupMaxSize)
   } else {
-    groups = kruskal(matrix, groupMaxSize)
+    groups = kruskal(matrix, groupMaxSize, threshold)
   }
   
   let ans: string[][] = [];
diff --git a/packages/visual-insights/src/insights/config.ts b/packages/visual-insights/src/insights/config.ts
index 45ba0a70..b486faa6 100644
--- a/packages/visual-insights/src/insights/config.ts
+++ b/packages/visual-insights/src/insights/config.ts
@@ -4,4 +4,11 @@ export const Depth = 4;
 
 export const VisualLimit = 8;
 
-export const TopKPercentField = 0.8; 
\ No newline at end of file
+export const TopKPercentField = 0.8;
+
+export const CrammersVThreshold = 0.3;
+
+export const PearsonCorrelation = {
+  strong: 0.5,
+  weak: 0.3
+};
\ No newline at end of file
diff --git a/packages/visual-insights/src/insights/impurity.ts b/packages/visual-insights/src/insights/impurity.ts
index 112359be..6c20ff87 100644
--- a/packages/visual-insights/src/insights/impurity.ts
+++ b/packages/visual-insights/src/insights/impurity.ts
@@ -2,6 +2,9 @@
 import aggregate from 'cube-core';
 import { entropy, normalize } from '../impurityMeasure';
 import { DataSource, OperatorType } from '../commonTypes';
+import { crammersV } from '../dashboard/utils';
+import { CrammersVThreshold } from './config';
+import cluster from './cluster';
 // insights like outlier and trend both request high impurity of dimension.
 const maxVisualChannel = 8;
 function getCombination(elements: string[], start: number = 1, end: number = elements.length): string[][] {
@@ -22,6 +25,36 @@ function getCombination(elements: string[], start: number = 1, end: number = ele
   }
   return ans
 }
+function getDimCorrelationMatrix(dataSource: DataSource, dimensions: string[]): number[][] {
+  let matrix: number[][] = dimensions.map(d => dimensions.map(d => 0));
+  for (let i = 0; i < dimensions.length; i++) {
+    matrix[i][i] = 1;
+    for(let j = i + 1; j < dimensions.length; j++) {
+      matrix[i][j] = matrix[j][i] = crammersV(dataSource, dimensions[i], dimensions[j]);
+    }
+  }
+  return matrix;
+}
+
+export function getDimSetsBasedOnClusterGroups(dataSource: DataSource, dimensions: string[]): string[][] {
+  const maxDimNumberInView = 4;
+  let dimSets: string[][] = [];
+  let dimCorrelationMatrix = getDimCorrelationMatrix(dataSource, dimensions);
+  console.log(dimCorrelationMatrix)
+  // groupMaxSize here means group number.
+  let groups: string[][] = cluster({
+    matrix: dimCorrelationMatrix,
+    measures: dimensions,
+    groupMaxSize: Math.round(dimensions.length / maxDimNumberInView),
+    threshold: CrammersVThreshold
+  });
+  // todo: maybe a threhold would be better ?
+  for (let group of groups) {
+    let combineDimSet: string[][] = getCombination(group);
+    dimSets.push(...combineDimSet);
+  }
+  return dimSets;
+}
 
 export function linearMapPositive (arr: number[]): number[] {
   let min = Math.min(...arr);
@@ -48,7 +81,7 @@ export function correlation(dataSource: DataSource, fieldX: string, fieldY: stri
 export type FieldsFeature = [string[], any, number[][]];
 function analysisDimensions(dataSource: DataSource, dimensions: string[], measures: string[], operator: OperatorType | undefined = 'sum'): FieldsFeature[] {
   let impurityList: FieldsFeature[] = [];
-  let dimSet = getCombination(dimensions)
+  let dimSet = getDimSetsBasedOnClusterGroups(dataSource, dimensions);
   for (let dset of dimSet) {
     let impurity = {};
     let aggData = aggregate({
diff --git a/packages/visual-insights/src/insights/index.ts b/packages/visual-insights/src/insights/index.ts
index d2e69085..4f6417d0 100644
--- a/packages/visual-insights/src/insights/index.ts
+++ b/packages/visual-insights/src/insights/index.ts
@@ -1,5 +1,5 @@
-import { analysisDimensions, getCombination } from './impurity';
-import { TopKSingleField, TopKPercentField, Depth, VisualLimit } from './config';
+import { analysisDimensions, getCombination, getDimSetsBasedOnClusterGroups } from './impurity';
+import { TopKPercentField } from './config';
 import { entropy, normalize } from '../impurityMeasure';
 import { memberCount } from '../utils'
 import cluster, { kruskalMST } from './cluster';
@@ -52,4 +52,10 @@ function getInsightViews(dataSource: DataSource, originDimensions: string[], mea
 }
 
 export default getInsightViews;
-export { analysisDimensions, getCombination, cluster as clusterMeasures, kruskalMST }
\ No newline at end of file
+export {
+  analysisDimensions,
+  getCombination,
+  getDimSetsBasedOnClusterGroups,
+  cluster as clusterMeasures,
+  kruskalMST
+};
\ No newline at end of file
diff --git a/packages/visual-insights/src/specification.ts b/packages/visual-insights/src/specification.ts
index 85b12604..60592735 100644
--- a/packages/visual-insights/src/specification.ts
+++ b/packages/visual-insights/src/specification.ts
@@ -1,11 +1,9 @@
 import { DataSource, FieldType, Field, FieldImpurity, Specification, View } from './commonTypes';
-import fieldsAnalysis from './fieldAnalysis';
 import {
   // isFieldCategory,
   // isFieldContinous,
   memberCount
 } from './utils';
-import { FieldSummary } from './univariateSummary';
 interface VisualElements {
   position: number;
   color: number;
diff --git a/packages/visual-insights/test/dashboard.js b/packages/visual-insights/test/dashboard.js
index 502bbe82..4fb45b5c 100644
--- a/packages/visual-insights/test/dashboard.js
+++ b/packages/visual-insights/test/dashboard.js
@@ -38,5 +38,29 @@ describe('insights test', function () {
     console.log(result)
     assert.equal(result.length, Math.pow(2, 6) - 1)
   })
+  /**
+   * test example
+   * https://www.empirical-methods.hslu.ch/decisiontree/relationship/chi-square-contingency/
+   */
+  it('test crammerV', function () {
+    let matrix = [
+      [19, 32, 83, 97, 48],
+      [2, 6, 16, 42, 26],
+      [0, 1, 3, 21, 10]
+    ]
+    let data = []
+    for (let i = 0; i < matrix.length; i++) {
+      for (let j = 0; j < matrix[i].length; j++) {
+        for (let k = 0; k < matrix[i][j]; k++) {
+          data.push({
+            x: 'x' + i,
+            y: 'y' + j
+          })
+        }
+      }
+    }
+    let result = DashBoard.crammersV(data, 'x', 'y');
+    assert.equal(result - 0.187 < 0.001, true);
+  })
 })
 
diff --git a/packages/visual-insights/test/insights.js b/packages/visual-insights/test/insights.js
index f10bf47c..50ff7127 100644
--- a/packages/visual-insights/test/insights.js
+++ b/packages/visual-insights/test/insights.js
@@ -2,7 +2,7 @@ const fs = require('fs');
 const assert = require('assert');
 const path  = require('path');
 
-const { analysisDimensions, Cleaner, getInsightViews, getCombination } = require('../build/cjs/index');
+const { analysisDimensions, Cleaner, getInsightViews, getCombination, getDimSetsBasedOnClusterGroups } = require('../build/cjs/index');
 
 const datasetPath = path.resolve(__dirname, './dataset/airbnb.json');
 const dataset = JSON.parse(fs.readFileSync(datasetPath).toString());
@@ -35,5 +35,12 @@ describe('insights test', function () {
     console.log(result)
     assert.equal(result.length, Math.pow(2, 6) - 1)
   })
+
+  it('print(clusterCombination vs. combination)', function () {
+    let result = getDimSetsBasedOnClusterGroups(cleanData, dimensions);
+    let unClusterResult = getCombination(dimensions);
+    console.log(result.length, unClusterResult.length, result)
+    assert.equal(result.length <= unClusterResult.length, true);
+  })
 })
 
diff --git a/packages/visual-insights/test/utils.js b/packages/visual-insights/test/utils.js
index c7648c04..492c2114 100644
--- a/packages/visual-insights/test/utils.js
+++ b/packages/visual-insights/test/utils.js
@@ -52,10 +52,10 @@ describe('utils test', function () {
     let t = true;
     for (let row of data) {
       if (typeof row[field] === 'number') {
-        let { groups: { left, right } } = /\[(?<left>([0-9.]+|-Infinity)), (?<right>[0-9.]+|Infinity)\)/.exec(row[newField]);
+        let { groups: { left, right } } = /.*\[(?<left>([0-9.e+]+|-Infinity)), (?<right>[0-9.e+]+|Infinity)\)/.exec(row[newField]);
           left = Number(left);
           right = Number(right);
-          if (!(left <= row[field] && row[field] < right)) {
+          if (!(left <= row[field] && row[field] <= right)) {
             t = false;
           }
       } else {