Chapter 9. Mining Complex Types
             Data Mining:                                                                 ...
A Travel Database for Plan Mining                                                                                         ...
Dimensions and Measures in
                   Spatial Data Warehousing                                                    ...
Spatial Association Analysis                                                               Progressive Refinement Mining o...
Approaches Based on Image
                    Signature                                                                   ...
Multidimensional Analysis                                                                                                 ...
Mining Multimedia Databases                                               Challenge: Curse of Dimensionality
 From Coarse ...
Estimation of Trend Curve                                                Discovery of Trend in Time-Series (1)

     n    ...
Subsequence Matching                                                       Enhanced similarity search methods

           ...
Sequential pattern mining: Cases and                                          Sequential pattern mining: Cases and
       ...
Information Retrieval                                                                  Basic Measures for Text Retrieval

...
Other Text Retrieval Indexing                                                          Types of Text Data Mining
         ...
Chapter 9. Mining Complex Types
             of Data                                                                      ...
Data Mining: Concepts and Techniques
Data Mining: Concepts and Techniques
Data Mining: Concepts and Techniques
Data Mining: Concepts and Techniques
Data Mining: Concepts and Techniques
Upcoming SlideShare
Loading in …5
×

Data Mining: Concepts and Techniques

3,192 views

Published on

Data Mining: Concepts and Techniques

  1. 1. Chapter 9. Mining Complex Types Data Mining: of Data Concepts and Techniques n Multidimensional analysis and descriptive mining of — Slides for Textbook — complex data objects — Chapter 9 — n Mining spatial databases n Mining multimedia databases ©Jiawei Han and Micheline Kamber n Mining time-series and sequence data Intelligent Database Systems Research Lab n Mining text databases School of Computing Science n Mining the World -Wide Web Simon Fraser University, Canada n Summary http://www.cs.sfu.ca January 17, 2001 Data Mining: Concepts and Techniques 1 January 17, 2001 Data Mining: Concepts and Techniques 2 Mining Complex Data Objects: Generalizing Spatial and Multimedia Data Generalization of Structured Data n Spatial data: n Set-valued attribute n Generalize detailed geographic points into clustered regions, n Generalization of each value in the set into its such as business, residential, industrial, or agricultural areas , corresponding higher-level concepts according to land usage n Derivation of the general behavior of the set, such n Require the merge of a set of geographic areas by spatial operations as the number of elements in the set, the types or value ranges in the set, or the weighted average n Image data: for numerical data n Extracted by aggregation and/or approximation n E.g., hobby = {tennis, hockey, chess, violin, n Size, color, shape, texture, orientation, and relative positions nintendo_games } generalizes to { sports, music, and structures of the contained objects or regions in the image video_games } n Music data: n List-valued or a sequence-valued attribute n Summarize its melody: based on the approximate patterns that n Same as set-valued attributes except that the order repeatedly occur in the segment of the elements in the sequence should be n Summarized its style: based on its tone, tempo, or the major observed in the generalization musical instruments played January 17, 2001 Data Mining: Concepts and Techniques 3 January 17, 2001 Data Mining: Concepts and Techniques 4 An Example: Plan Mining by Divide and Generalizing Object Data Conquer n Object identifier: generalize to the lowest level of class in the n Plan: a variable sequence of actions class/subclass hierarchies n Class composition hierarchies n E.g., Travel (flight): <traveler, departure, arrival, d-time, a-time, n generalize nested structured data airline, price, seat> n generalize only objects closely related in semantics to the current n Plan mining: extraction of important or significant generalized one (sequential) patterns from a planbase (a large collection of plans) n Construction and mining of object cubes n Extend the attribute-oriented induction method n E.g., Discover travel patterns in an air flight database, or n Apply a sequence of class -based generalization operators on n find significant patterns from the sequences of actions in the different attributes repair of automobiles n Continue until getting a small number of generalized objects that n Method can be summarized as a concise in high-level terms n For efficient implementation n Attribute-oriented induction on sequence data n Examine each attribute, generalize it to simple-valued data n A generalized travel plan: <small -big*-small> n Construct a multidimensional data cube (object cube) n Divide & conquer:Mine characteristics for each subsequence n Problem: it is not always desirable to generalize a set of values n E.g., big*: same airline, small big: nearby region - to single-valued data January 17, 2001 Data Mining: Concepts and Techniques 5 January 17, 2001 Data Mining: Concepts and Techniques 6 1
  2. 2. A Travel Database for Plan Mining Multidimensional Analysis n Example: Mining a travel planbase n Strategy A multi-D model for the planbase Travel plans table plan# action# departure depart_time arrival arrival_time airline … n Generalize the 1 1 1 2 ALB JFK 800 1000 JFK ORD 900 1230 TWA UA … … planbase in 1 1 3 4 ORD LAX 1300 1710 LAX SAN 1600 1800 UA DAL … … different 2 1 SPI 900 ORD 950 AA … directions . . . . . . . . . . . . . . . . . . . . . . . . n Look for sequential Airport info table airport_code city state region airport_size … patterns in the 1 1 1 2 ALB JFK 800 1000 … … generalized plans Derive high-level 1 3 ORD 1300 … 1 4 LAX 1710 … n plans 2 1 SPI 900 … . . . . . . . . . . . . . . . January 17, 2001 Data Mining: Concepts and Techniques 7 January 17, 2001 Data Mining: Concepts and Techniques 8 Multidimensional Generalization Generalization-Based Sequence Mining Multi-D generalization of the planbase Plan# Loc_Seq Size_Seq State_Seq 1 ALB - JFK - ORD - LAX - SAN S-L-L-L-S N-N-I-C-C n Generalize planbase in multidimensional way using 2 SPI - ORD - JFK - SYR S-L-L-S I-I-N-N . . . . . . dimension tables . . . Merging consecutive, identical actions in plans n Use # of distinct values (cardinality) at each level to Plan# Size_Seq State_Seq Region_Seq … determine the right level of generalization (level- 1 2 S - L+ - S S - L+ - S N+ - I - C+ I+ - N+ E+ - M - P+ M+ - E+ … … “planning”) . . . . . . n Use operators merge “+”, option “[]” to further . . . generalize patterns flight( x, y , ) ∧ airport _ size( x, S ) ∧ airport _ size( y, L) n Retain patterns with significant support ⇒ region( x) = region( y ) [ 75%] January 17, 2001 Data Mining: Concepts and Techniques 9 January 17, 2001 Data Mining: Concepts and Techniques 10 Chapter 9. Mining Complex Types Generalized Sequence Patterns of Data n AirportSize-sequence survives the min threshold (after n Multidimensional analysis and descriptive mining of applying merge operator): complex data objects S-L+ -S [35%], L+ -S [30%], S-L+ [24.5%], L+ [9%] n Mining spatial databases n After applying option operator: n Mining multimedia databases [S] -L+ -[S] [98.5%] n Mining time-series and sequence data n Most of the time, people fly via large airports to get to n Mining text databases final destination n Mining the World -Wide Web n Other plans: 1.5% of chances, there are other patterns: S-S, L-S-L n Summary January 17, 2001 Data Mining: Concepts and Techniques 11 January 17, 2001 Data Mining: Concepts and Techniques 12 2
  3. 3. Dimensions and Measures in Spatial Data Warehousing Spatial Data Warehouse n Spatial data warehouse: Integrated, subject-oriented, n Dimension modeling n Measures time-variant, and nonvolatile spatial data repository for n nonspatial n numerical data analysis and decision making n e.g. temperature: 25-30 distributive (e.g. count, degrees generalizes to n n Spatial data integration: a big issue sum) hot n Structure -specific formats (raster- vs. vector-based, n spatial-to-nonspatial n algebraic (e.g. average) OO vs. relational models, different storage and n e.g. region “B.C.” holistic (e.g. median, indexing, etc.) n generalizes to rank) n Vendor-specific formats (ESRI, MapInfo, Integraph, description “western etc.) provinces” n spatial n spatial-to-spatial collection of spatial n Spatial data cube: multidimensional spatial database n e.g. region “Burnaby” n pointers (e.g. pointers to n Both dimensions and measures may contain spatial generalizes to region all regions with 25-30 components “Lower Mainland” degrees in July) January 17, 2001 Data Mining: Concepts and Techniques 13 January 17, 2001 Data Mining: Concepts and Techniques 14 Star Schema of the BC Weather Example: BC weather pattern analysis Warehouse n Input n Spatial data warehouse n A map with about 3,000 weather probes scattered in B.C. n Daily data for temperature, precipitation, wind velocity, etc. n Dimensions n Concept hierarchies for all attributes n region_name n Output n time n A map that reveals patterns: merged (similar) regions n temperature n Goals n precipitation n Interactive analysis (drill-down, slice, dice, pivot, roll-up) n Fast response time n Measurements n Minimizing storage space used n region_map n Challenge n area n A merged region may contain hundreds of “primitive” regions n count (polygons) Dimension table Fact table January 17, 2001 Data Mining: Concepts and Techniques 15 January 17, 2001 Data Mining: Concepts and Techniques 16 Methods for Computation of Spatial Merge Spatial Data Cube è Precomputing all: too n O n-line aggregation: collect and store pointers to spatial much storage space objects in a spatial data cube è On-line merge: very n expensive and slow, need efficient aggregation expensive techniques n Precompute and store all the possible combinations n huge space overhead n Precompute and store rough approximations in a spatial data cube n accuracy trade-off n Selective computation: only materialize those which will be accessed frequently n a reasonable choice January 17, 2001 Data Mining: Concepts and Techniques 17 January 17, 2001 Data Mining: Concepts and Techniques 18 3
  4. 4. Spatial Association Analysis Progressive Refinement Mining of Spatial Association Rules n Spatial association rule: A ⇒ B [s%, c% ] n Hierarchy of spatial relationship: n A and B are sets of spatial or nonspatial predicates n g_close_to: near_by , touch, intersect, contain, etc. n Topological relations: intersects, overlaps, disjoint, etc. n First search for rough relationship and then refine it n Spatial orientations: left_of, west_of, under, etc. n Two-step mining of spatial association: n Distance information: close_to, within_distance, etc. n Step 1: Rough spatial computation (as a filter) n s% is the support and c% is the confidence of the rule n Using MBR or R-tree for rough estimation n Step2: Detailed spatial algorithm (as refinement) n Examples n Apply only to those objects which have passed the rough is_a(x, large_town) ^ intersect(x, highway) → adjacent_to(x, water) spatial association test (no less than min_support ) [7%, 85%] is_a(x, large_town) ^adjacent_to(x, georgia_strait) → close_to(x, u.s.a.) [1%, 78%] January 17, 2001 Data Mining: Concepts and Techniques 19 January 17, 2001 Data Mining: Concepts and Techniques 20 Spatial Classification and Spatial Chapter 9. Mining Complex Types Trend Analysis of Data n Spatial classification n Analyze spatial objects to derive classification n Multidimensional analysis and descriptive mining of schemes, such as decision trees in relevance to certain complex data objects spatial properties (district, highway, river, etc.) n Example: Classify regions in a province into rich vs. n Mining spatial databases poor according to the average family income n Mining multimedia databases n Spatial trend analysis n Detect changes and trends along a spatial dimension n Mining time-series and sequence data n Study the trend of nonspatial or spatial data changing n Mining text databases with space n Example: Observe the trend of changes of the climate n Mining the World -Wide Web or vegetation with the increasing distance from an n Summary ocean January 17, 2001 Data Mining: Concepts and Techniques 21 January 17, 2001 Data Mining: Concepts and Techniques 22 Queries in Content-Based Similarity Search in Multimedia Data Retrieval Systems n Description-based retrieval systems n Image sample-based queries: n Find all of the images that are similar to the given n Build indices and perform object retrieval based on image sample image descriptions, such as keywords, captions, size, n Compare the feature vector (signature) extracted from and time of creation the sample with the feature vectors of images that n Labor -intensive if performed manually have already been extracted and indexed in the image database n Results are typically of poor quality if automated n Image feature specification queries: n Content-based retrieval systems n Specify or sketch image features like color, texture, or n Support retrieval based on the image content, such shape, which are translated into a feature vector as color histogram, texture, shape, objects, and n Match the feature vector with the feature vectors of wavelet transforms the images in the database January 17, 2001 Data Mining: Concepts and Techniques 23 January 17, 2001 Data Mining: Concepts and Techniques 24 4
  5. 5. Approaches Based on Image Signature Wavelet Analysis n Color histogram-based signature n Wavelet-based signature n Use the dominant wavelet coefficients of an image as its n The signature includes color histograms based on color signature composition of an image regardless of its scale or n Wavelets capture shape, texture, and location orientation information in a single unified framework n No information about shape, location, or texture n Improved efficiency and reduced the need for providing n Two images with similar color composition may contain multiple search primitives n May fail to identify images containing similar in location very different shapes or textures, and thus could be completely unrelated in semantics or size objects n Wavelet-based signature with region-based granularity n Multifeature composed signature n Similar images may contain similar regions, but a region n The signature includes a composition of multiple in one image could be a translation or scaling of a features: color histogram, shape, location, and texture matching region in the other n Compute and compare signatures at the granularity of n Can be used to search for similar images regions, not the entire image January 17, 2001 Data Mining: Concepts and Techniques 25 January 17, 2001 Data Mining: Concepts and Techniques 26 C-BIRD: Content -Based Image Multi-Dimensional Search in Retrieval from Digital libraries Multimedia Databases Color layout Search n by image colors n by color percentage n by color layout n by texture density n by texture Layout n by object model nby illumination invariance n by keywords January 17, 2001 Data Mining: Concepts and Techniques 27 January 17, 2001 Data Mining: Concepts and Techniques 28 Multi-Dimensional Analysis in Multimedia Databases Mining Multimedia Databases Color histogram Texture layout Refining or combining searches Search for “airplane in blue sky” (top layout grid is blue and keyword = “airplane”) Search for “blue sky and green meadows” Search for “blue sky” (top layout grid is blue (top layout grid is blue) and bottom is green) January 17, 2001 Data Mining: Concepts and Techniques 29 January 17, 2001 Data Mining: Concepts and Techniques 30 5
  6. 6. Multidimensional Analysis Mining Multimedia Databases in of Multimedia Data n Multimedia data cube n Design and construction similar to that of traditional data cubes from relational data n Contain additional dimensions and measures for multimedia information, such as color, texture, and shape n The database does not store images but their descriptors n Feature descriptor: a set of vectors for each visual characteristic n Color vector: contains the color histogram n MFC (Most Frequent Color) vector: five color centroids n MFO (Most Frequent Orientation) vector: five edge orientation centroids n Layout descriptor : contains a color layout vector and an edge layout vector January 17, 2001 Data Mining: Concepts and Techniques 31 January 17, 2001 Data Mining: Concepts and Techniques 32 Mining Multimedia Databases Classification in MultiMediaMiner The Data Cube and the Sub-Space Measurements JP GIF l EG Smal ium Med ge Large Lar ery V By Size By Format By Format & Size RED WHITE BLUE Cross Tab By Colour & Size JPEG GIF By Colour By Format & Colour RED WHITE Sum By Colour BLUE • Format of image By Format • Duration Group By Sum Colour • Colors RED • Textures WHITE • Keywords BLUE • Size Measurement • Width Sum • Height • Internet domain of image • Internet domain of parent pages • Image popularity January 17, 2001 Data Mining: Concepts and Techniques 33 January 17, 2001 Data Mining: Concepts and Techniques 34 Mining Associations in Multimedia Data Mining Multimedia Databases n Special features: Spatial Relationships from Layout n Need # of occurrences besides Boolean existence, e.g., property P1 on-top-of property P2 property P1 next-to property P2 n “Two red square and one blue circle” implies theme “air-show” n Need spatial relationships n Blue on top of white squared object is associated with brown bottom n Need multi-resolution and progressive refinement mining Different Resolution Hierarchy n It is expensive to explore detailed associations among objects at high resolution n It is crucial to ensure the completeness of search at multi-resolution space January 17, 2001 Data Mining: Concepts and Techniques 35 January 17, 2001 Data Mining: Concepts and Techniques 36 6
  7. 7. Mining Multimedia Databases Challenge: Curse of Dimensionality From Coarse to Fine Resolution Mining n Difficult to implement a data cube efficiently given a large number of dimensions, especially serious in the case of multimedia data cubes n Many of these attributes are set-oriented instead of single-valued n Restricting number of dimensions may lead to the modeling of an image at a rather rough, limited, and imprecise scale n More research is needed to strike a balance between efficiency and power of representation January 17, 2001 Data Mining: Concepts and Techniques 37 January 17, 2001 Data Mining: Concepts and Techniques 38 Chapter 9. Mining Complex Types Mining Time-Series and Sequence of Data Data n Time-series database n Multidimensional analysis and descriptive mining of n Consists of sequences of values or events changing complex data objects with time n Mining spatial databases n Data is recorded at regular intervals n Mining multimedia databases n Characteristic time-series components n Mining time-series and sequence data n Trend, cycle, seasonal, irregular n Mining text databases n Applications n Mining the World -Wide Web n Financial: stock price, inflation n Biomedical: blood pressure n Summary n Meteorological: precipitation January 17, 2001 Data Mining: Concepts and Techniques 39 January 17, 2001 Data Mining: Concepts and Techniques 40 Mining Time-Series and Sequence Mining Time-Series and Sequence Data Data: Trend analysis n A time series can be illustrated as a time-series graph Time-series plot which describes a point moving with the passage of time n Categories of Time -Series Movements n Long-term or trend movements (trend curve) n Cyclic movements or cycle variations, e.g., business cycles n Seasonal movements or seasonal variations n i.e, almost identical patterns that a time series appears to follow during corresponding months of successive years. n Irregular or random movements January 17, 2001 Data Mining: Concepts and Techniques 41 January 17, 2001 Data Mining: Concepts and Techniques 42 7
  8. 8. Estimation of Trend Curve Discovery of Trend in Time-Series (1) n The freehand method n Estimation of seasonal variations n Fit the curve by looking at the graph n Seasonal index nCostly and barely reliable for large-scaled data mining n Set of numbers showing the relative values of a variable during n The least-square method the months of the year E.g., if the sales during October, November, and December are nFind the curve minimizing the sum of the squares of n 80%, 120%, and 140% of the average monthly sales for the the deviation of points on the curve from the whole year, respectively, then 80, 120, and 140 are seasonal corresponding data points index numbers for these months n The moving-average method n Deseasonalized data n Eliminate cyclic, seasonal and irregular patterns n Data adjusted for seasonal variations n Loss of end data n E.g., divide the original monthly data by the seasonal index n Sensitive to outliers numbers for the corresponding months January 17, 2001 Data Mining: Concepts and Techniques 43 January 17, 2001 Data Mining: Concepts and Techniques 44 Discovery of Trend in Time-Series (2) Similarity Search in Time-Series Analysis n Normal database query finds exact match n Estimation of cyclic variations n Similarity search finds data sequences that differ only n If (approximate) periodicity of cycles occurs, cyclic slightly from the given query sequence index can be constructed in much the same manner n Two categories of similarity queries as seasonal indexes n Whole matching: find a sequence that is similar to the n Estimation of irregular variations query sequence n Subsequence matching: find all pairs of similar n By adjusting the data for trend, seasonal and cyclic sequences variations n Typical Applications n With the systematic analysis of the trend, cyclic, n Financial market seasonal, and irregular components, it is possible to n Market basket data analysis make long- or short -term predictions with reasonable n Scientific databases quality n Medical diagnosis January 17, 2001 Data Mining: Concepts and Techniques 45 January 17, 2001 Data Mining: Concepts and Techniques 46 Data transformation Multidimensional Indexing n Many techniques for signal analysis require the data to n Multidimensional index be in the frequency domain n Constructed for efficient accessing using the first few n Usually data-independent transformations are used Fourier coefficients n The transformation matrix is determined a priori n Use the index can to retrieve the sequences that are at n E.g., discrete Fourier transform (DFT), discrete most a certain small distance away from the query wavelet transform (DWT) sequence n The distance between two signals in the time domain n Perform postprocessing by computing the actual is the same as their Euclidean distance in the distance between sequences in the time domain and frequency domain discard any false matches n DFT does a good job of concentrating energy in the first few coefficients n If we keep only first a few coefficients in DFT, we can compute the lower bounds of the actual distance January 17, 2001 Data Mining: Concepts and Techniques 47 January 17, 2001 Data Mining: Concepts and Techniques 48 8
  9. 9. Subsequence Matching Enhanced similarity search methods n Allow for gaps within a sequence or differences in offsets n Break each sequence into a set of pieces of window with length w or amplitudes n Normalize sequences with amplitude scaling and offset n Extract the features of the subsequence inside the window translation n Map each sequence to a “trail” in the feature space n Two subsequences are considered similar if one lies n Divide the trail of each sequence into “subtrails” and within an envelope of ε width around the other, ignoring represent each of them with minimum bounding rectangle outliers n Use a multipiece assembly algorithm to search for longer n Two sequences are said to be similar if they have enough sequence matches non-overlapping time-ordered pairs of similar subsequences n Parameters specified by a user or expert: sliding window size, width of an envelope for similarity, maximum gap, and matching fraction January 17, 2001 Data Mining: Concepts and Techniques 49 January 17, 2001 Data Mining: Concepts and Techniques 50 Steps for performing a similarity Query Languages for Time Sequences search n Atomic matching n Time-sequence query language n Find all pairs of gap-free windows of a small length n Should be able to specify sophisticated queries like that are similar Find all of the sequences that are similar to some sequence in class A, but not similar to any sequence in class B n Window stitching Should be able to support various kinds of queries: range n Stitch similar windows to form pairs of large similar n queries, all-pair queries, and nearest neighbor queries subsequences allowing gaps between atomic matches n Shape definition language Allows users to define and query the overall shape of time n Subsequence Ordering n sequences n Linearly order the subsequence matches to n Uses human readable series of sequence transitions or macros determine whether enough similar pieces exist n Ignores the specific details n E.g., the pattern up, Up, UP can be used to describe increasing degrees of rising slopes n Macros: spike, valley, etc. January 17, 2001 Data Mining: Concepts and Techniques 51 January 17, 2001 Data Mining: Concepts and Techniques 52 Sequential Pattern Mining Mining Sequences (cont.) n Mining of frequently occurring patterns related to time or Customer-sequence Map Large Itemsets other sequences n Sequential pattern mining usually concentrate on symbolic Large Itemsets MappedID patterns CustId Video sequence (C) 1 n Examples 1 {(C), (H)} (D) 2 2 {(AB), (C), (DFG)} (G) 3 n Renting “Star Wars”, then “Empire Strikes Back”, then 3 {(CEG)} “Return of the Jedi” in that order (DG) 4 4 {(C), (DG), (H)} (H) 5 n Collection of ordered events within an interval 5 {(H)} n Applications Sequential patterns with support > 0.25 n Targeted marketing {(C), (H)} n Customer retention {(C), (DG)} n Weather prediction January 17, 2001 Data Mining: Concepts and Techniques 53 January 17, 2001 Data Mining: Concepts and Techniques 54 9
  10. 10. Sequential pattern mining: Cases and Sequential pattern mining: Cases and Parameters Parameters (2) n Duration of a time sequence T n Time interval, int, between events in the discovered n Sequential pattern mining can then be confined to the pattern data within a specified duration n Ex. Subsequence corresponding to the year of 1999 n int = 0: no interval gap is allowed, i.e., only strictly consecutive sequences are found n Ex. Partitioned sequences, such as every year, or every Ex. “Find frequent patterns occurring in consecutive weeks” week after stock crashes, or every two weeks before n and after a volcano eruption n min_int ≤ int ≤ max_int: find patterns that are n Event folding window w separated by at least min_int but at most max_int n If w = T, time -insensitive frequent patterns are found n Ex. “If a person rents movie A, it is likely she will rent movie B within 30 days” (int ≤ 30) n If w = 0 (no event sequence folding), sequential patterns are found where each event occurs at a n int = c ≠ 0: find patterns carrying an exact interval distinct time instant n Ex. “Every time when Dow Jones drops more than 5%, what will happen exactly two days later?” (int = 2) n If 0 < w < T, sequences occurring within the same period w are folded in the analysis January 17, 2001 Data Mining: Concepts and Techniques 55 January 17, 2001 Data Mining: Concepts and Techniques 56 Episodes and Sequential Pattern Periodicity Analysis Mining Methods n Periodicity is everywhere: tides, seasons, daily power n Other methods for specifying the kinds of patterns consumption, etc. n Full periodicity n Serial episodes: A → B n Every point in time contributes (precisely or n Parallel episodes: A & B approximately) to the periodicity n Partial periodicit: A more general notion n Regular expressions: (A | B)C*(D → E) n Only some segments contribute to the periodicity n Methods for sequential pattern mining n Jim reads NY Times 7:00-7:30 am every week day n Variations of Apriori-like algorithms, e.g., GSP n Cyclic association rules n Associations which form cycles n Database projection-based pattern growth n Methods n Similar to the frequent pattern growth without n Full periodicity: FFT, other statistical analysis methods candidate generation n Partial and cyclic periodicity: Variations of Apriori-like mining methods January 17, 2001 Data Mining: Concepts and Techniques 57 January 17, 2001 Data Mining: Concepts and Techniques 58 Chapter 9. Mining Complex Types Text Databases and IR of Data n Text databases (document databases) n Multidimensional analysis and descriptive mining of n Large collections of documents from various sources: news articles, research papers, books, digital libraries, complex data objects e-mail messages, and Web pages, library database, etc. n Data stored is usually semi-structured n Mining spatial databases n Traditional information retrieval techniques become n Mining multimedia databases inadequate for the increasingly vast amounts of text data n Mining time-series and sequence data n Information retrieval n A field developed in parallel with database systems n Mining text databases n Information is organized into (a large number of) n Mining the World -Wide Web documents n Information retrieval problem: locating relevant n Summary documents based on user input, such as keywords or example documents January 17, 2001 Data Mining: Concepts and Techniques 59 January 17, 2001 Data Mining: Concepts and Techniques 60 10
  11. 11. Information Retrieval Basic Measures for Text Retrieval n Typical IR systems n Online library catalogs n Online document management systems n Precision: the percentage of retrieved documents that are n Information retrieval vs. database systems in fact relevant to the query (i.e., “correct” responses) n Some DB problems are not present in IR, e.g., update, | {Relevant} ∩{ Retrieved} | precision= transaction management, complex objects | { Retrieved} | n Some IR problems are not addressed well in DBMS, n Recall: the percentage of documents that are relevant to e.g., unstructured documents, approximate search the query and were, in fact, retrieved | {Relevant} ∩ {Retrieved} | using keywords and relevance precision= |{ Relevant} | January 17, 2001 Data Mining: Concepts and Techniques 61 January 17, 2001 Data Mining: Concepts and Techniques 62 Keyword-Based Retrieval Similarity-Based Retrieval in Text Databases n A document is represented by a string, which can be n Finds similar documents based on a set of common identified by a set of keywords keywords n Queries may use expressions of keywords n Answer should be based on the degree of relevance n E.g., car and repair shop, tea or coffee, DBMS but based on the nearness of the keywords, relative not Oracle frequency of the keywords, etc. n Queries and retrieval should consider synonyms, e.g., repair and maintenance n Basic techniques n Major difficulties of the model n Stop list n Synonymy : A keyword T does not appear anywhere n Set of words that are deemed “irrelevant”, even in the document, even though the document is though they may appear frequently closely related to T, e.g., data mining n E.g., a, the, of, for, with, etc. n Polysemy : The same keyword may mean different n Stop lists may vary when document set varies things in different contexts, e.g., mining January 17, 2001 Data Mining: Concepts and Techniques 63 January 17, 2001 Data Mining: Concepts and Techniques 64 Similarity-Based Retrieval in Text Databases (2) Latent Semantic Indexing n Word stem n Basic idea n Several words are small syntactic variants of each n Similar documents have similar word frequencies other since they share a common word stem n Difficulty: the size of the term frequency matrix is very large n E.g., drug, drugs, drugged n Use a singular value decomposition (SVD) techniques to reduce the size of frequency table n A term frequency table n Retain the K most significant rows of the frequency table n Each entry frequent_table(i, j) = # of occurrences n Method of the word t i in document di n Create a term frequency matrix, freq_matrix n Usually, the ratio instead of the absolute number of SVD construction: Compute the singular valued decomposition of occurrences is used n freq_matrix by splitting it into 3 matrices, U, S, V n Similarity metrics: measure the closeness of a document n Vector identification: For each document d, replace its original to a query (a set of keywords) document vector by a new excluding the eliminated terms n Relative term occurrences Index creation: Store the set of all vectors, indexed by one of a v ⋅v n n Cosine distance: sim(v1, v2 ) = 1 2 number of techniques (such as TV-tree) | v1 | | v2 | January 17, 2001 Data Mining: Concepts and Techniques 65 January 17, 2001 Data Mining: Concepts and Techniques 66 11
  12. 12. Other Text Retrieval Indexing Types of Text Data Mining Techniques n Inverted index n Keyword-based association analysis n Maintains two hash- or B+-tree indexed tables: n Automatic document classification n document_table: a set of document records <doc_id, n Similarity detection postings_list> n Cluster documents by a common author n term_table: a set of term records, <term, postings_list> n Cluster documents containing information from a n Answer query: Find all docs associated with one or a set of terms common source Advantage: easy to implement Link analysis: unusual correlation between entities n n n Disadvantage: do not handle well synonymy and polysemy, and posting lists could be too long (storage could be very large) n Sequence analysis: predicting a recurring event n Signature file n Anomaly detection: find information that violates usual n Associate a signature with each document patterns n A signature is a representation of an ordered list of terms that n Hypertext analysis describe the document n Patterns in anchors/links n Order is obtained by frequency analysis, stemming and stop lists n Anchor text correlations with linked objects January 17, 2001 Data Mining: Concepts and Techniques 67 January 17, 2001 Data Mining: Concepts and Techniques 68 Keyword-based association analysis Automatic document classification n Collect sets of keywords or terms that occur frequently n Motivation together and then find the association or correlation n Automatic classification for the tremendous number of relationships among them on-line text documents (Web pages, e-mails, etc.) n First preprocess the text data by parsing, stemming, n A classification problem removing stop words, etc. n Training set: Human experts generate a training data set n Then evoke association mining algorithms n Classification: The computer system discovers the n Consider each document as a transaction classification rules n View a set of keywords in the document as a set of n Application: The discovered rules can be applied to items in the transaction classify new/unknown documents n Term level association mining n Text document classification differs from the classification of n No need for human effort in tagging documents relational data n The number of meaningless results and the execution n Document databases are not structured according to time is greatly reduced attribute-value pairs January 17, 2001 Data Mining: Concepts and Techniques 69 January 17, 2001 Data Mining: Concepts and Techniques 70 Association-Based Document Document Clustering Classification n Extract keywords and terms by information retrieval and simple association analysis techniques n Automatically group related documents based on their n Obtain concept hierarchies of keywords and terms using contents n Available term classes, such as WordNet n Require no training sets or predetermined taxonomies, n Expert knowledge generate a taxonomy at runtime n Some keyword classification systems n Major steps n Preprocessing n Classify documents in the training set into class hierarchies n Remove stop words, stem, feature extraction, lexical n Apply term association mining method to discover sets of associated terms analysis, … n Use the terms to maximally distinguish one class of documents from n Hierarchical clustering others n Compute similarities applying clustering algorithms, n Derive a set of association rules associated with each document class … n Order the classification rules based on their occurrence frequency n Slicing and discriminative power n Fan out controls, flatten the tree to configurable n Used the rules to classify new documents number of levels, … January 17, 2001 Data Mining: Concepts and Techniques 71 January 17, 2001 Data Mining: Concepts and Techniques 72 12
  13. 13. Chapter 9. Mining Complex Types of Data Mining the World-Wide Web n The WWW is huge, widely distributed, global n Multidimensional analysis and descriptive mining of information service center for n Information services: news, advertisements, complex data objects consumer information, financial management, n Mining spatial databases education, government, e-commerce, etc. n Hyper-link information n Mining multimedia databases n Access and usage information n Mining time-series and sequence data n WWW provides rich sources for data mining n Mining text databases n Challenges n Too huge for effective data warehousing and data n Mining the World -Wide Web mining n Summary n Too complex and heterogeneous: no standards and structure January 17, 2001 Data Mining: Concepts and Techniques 73 January 17, 2001 Data Mining: Concepts and Techniques 74 Mining the World-Wide Web Web search engines n Growing and changing very rapidly Internet growth n Index-based: search the Web, index Web pages, and 40000000 35000000 build and store huge keyword-based indices 30000000 25000000 n Help locate sets of Web pages containing certain Hosts 20000000 15000000 keywords 10000000 5000000 0 n Deficiencies n A topic of any breadth may easily contain hundreds of Sep-69 Sep-72 Sep-75 Sep-78 Sep-81 Sep-84 Sep-87 Sep-90 Sep-93 Sep-96 Sep-99 thousands of documents n Broad diversity of user communities n Only a small portion of the information on the Web is truly relevant or n Many documents that are highly relevant to a topic useful may not contain keywords defining them (polysemy ) n 99% of the Web information is useless to 99% of Web users n How can we find high-quality Web pages on a specified topic? January 17, 2001 Data Mining: Concepts and Techniques 75 January 17, 2001 Data Mining: Concepts and Techniques 76 Web Mining: A more challenging task Web Mining Taxonomy n Searches for n Web access patterns n Web structures Web Mining n Regularity and dynamics of Web contents n Problems Web Content Web Structure Web Usage n The “abundance” problem Mining Mining Mining n Limited coverage of the Web: hidden Web sources, majority of data in DBMS Web Page Content Mining Search Result Mining General Access Pattern Tracking Customized Usage Tracking n Limited query interface based on keyword-oriented search n Limited customization to individual users January 17, 2001 Data Mining: Concepts and Techniques 77 January 17, 2001 Data Mining: Concepts and Techniques 78 13

×