UNIT-II

2.1 INTRODUCTION TO STREAMS CONCEPTS

Stream processing is a data management technique that involves ingesting a continuous data
stream to quickly analyze, filter, transform or enhance the data in real time.

Big data streaming is a process in which big data is quickly processed in order to extract real-
time insights from it. Big data streaming is ideally a speed-focused approach wherein a
continuous stream of data is processed.

Big data streaming is a process in which large streams of real-time data are processed with the
sole aim of extracting insights and useful trends out of it. A continuous stream of unstructured
data is sent for analysis into memory before storing it onto disk. This happens across a cluster
of servers. Speed matters the most in big data streaming. The value of data, if not processed
quickly, decreases with time.

Streams: Temporally ordered, fast changing, massive and potentially infinite

Data Stream:

 A data stream is a Massive sequence of data

 Too large to store (on disk, memory, cache etc.,)

Types of Data Streams:

Data stream –
A data stream is a (possibly unchained) sequence of tuples. Each tuple comprised of a set
of attributes, similar to a row in a database table.
 Transactional data stream –
It is a log interconnection between entities
 Credit card – purchases by consumers from producer
 Telecommunications – phone calls by callers to the dialed parties
 Web – accesses by clients of information at servers
 Measurement data streams –
 Sensor Networks – a physical natural phenomenon, road traffic
 IP Network – traffic at router interfaces
 Earth climate – temperature, humidity level at weather stations

Characteristics of Data Streams:

 Large volumes of continuous data, possibly infinite.
 Steady changing and requires a fast, real-time response.
 Data stream captures nicely our data processing needs of today.
 Random access is expensive and a single scan algorithm
 Store only the summary of the data seen so far.
 Maximum stream data are at a pretty low level or multidimensional in creation, needs
multilevel and multidimensional treatment.
Challenges in Working with Streaming Data

o Storage layer
o Processing Layer
o Scalability
o Data Durability
o Fault Tolerance in both the Storage and processing layer.

Stream Data Sources

 Sensor Data
 Image Data
 Internet and Web Traffic

BATCH PROCESSING VS. REAL-TIME STREAM PROCESSING

In batch data processing, data is downloaded in batches before being processed, stored, and
analyzed. On the other hand, stream data ingest data continuously, allowing it to be processed
simultaneously and in real-time.

The main benefit of stream processing is real-time insight. We live in an information age where
new data is constantly being created. Organizations that leverage streaming data analytics can
take advantage of real-time information from internal and external assets to inform their
decisions, drive innovation and improve their overall strategy.
 2.2 STREAM DATA MODEL AND ARCHITECTURE

Any number of streams can enter the system. Each stream can provide elements at its own
schedule; they need not have the same data rates or data types, and the time between elements
of one stream need not be uniform. The fact that the rate of arrival of stream elements is not
under the control of the system distinguishes stream processing from the processing of data
that goes on within a database-management system. The latter system controls the rate at which
data is read from the disk, and therefore never has to worry about data getting lost as it attempts
to execute queries.

 Data Stream Management System

 It is a real time continuous, ordered (implicitly by arrival time and explicitly by Timestamp)
sequence if items.
 Hard to control the order in which items arrive
 It is not feasible to locally store a stream in its entirety.
 DSMS(Data stream management system) is a computer program that permits to manage
continuous data streams
 Stream Processor
Stream processor receives the data stream
 Data Streams are volatile
 Provides sequential access to data.
 Stream Data changes continuously
 Data streams are generated by various sources
 Data can be viewed as an infinite, time oriented sequence of tuples.
 Any number of streams can enter the system
 Standing query
User need to write Queries and are need to place in processor to process data stream and store
the results in limited working storage (either Main memory/ Disk). These queries are, in a
sense, permanently executing, and produce outputs at appropriate times. Standing queries are
designed to answer Ad-hoc queries.
Example:
1. Standing query to output an alert whenever the temperature exceeds 25 degrees
centigrade from the stream produced by the ocean-surface-temperature sensor.
2. If we want the average temperature over all time, we have only to record two values:
the number of readings ever sent in the stream and the sum of those readings
 Ad-Hoc Queries
If we want the facility to ask a wide variety of ad-hoc queries, a common approach is to store
a sliding window of each stream in the working store. A sliding window can be the most
recent n elements of a stream, for some n, or it can be all the elements that arrived within the
last t time units, e.g., one day.

 Archival Store
Streams may be archived in a large archival store, but we assume it is not possible to answer
queries from the archival store. It could be examined only under special circumstances using
time-consuming retrieval processes.

 Working Store
There is also a working store, into which summaries or parts of streams may be placed, and
which can be used for answering queries. The working store might be disk, or it might be main
memory, depending on how fast we need to process queries.
But either way, it is of sufficiently limited capacity that it cannot store all the data from
all the streams.

Benefits of Stream Data Model

 Able to deal with never-ending streams of events

 Real-time or near-real-time processing
 Detecting patterns in time-series data
 Easy Data scalability

Types of queries on Data stream

o Filtering a Data stream

- Select element with property x from the stream
o Counting distinct elements
- Number of distinct elements from the last k elements of the stream
o Estimating moments
- Estimating Avg./ Std. dev. of last k elements.
o Finding frequent elements
- Identifying which element is repeatedly coming
-

2.3 STREAM COMPUTING

Stream computing is a way to analyze and process Big Data in real time to gain current insights
to take appropriate decisions or to predict new trends in the immediate future. Streams are High
rate of receiving data and are Implements in a distributed clustered environment.

Stream computing Applications

o Financial sectors
o Business intelligence
o Risk management
 o Marketing management
o Search engines
o Social network analysis
o Mining query streams
 Ex: Google wants to know what queries are more frequent today than yesterday
o Mining click streams
 Ex: Yahoo wants to know which of its pages are getting an unusual number of hits
in the past hour
o Mining social network news feeds
 E.g., Look for trending topics on Twitter, Facebook

2.4 SAMPLING DATA IN A STREAM

The general problem we shall address is selecting a subset of a stream so that we can ask queries
about the selected subset and have the answers be statistically representative of the stream as a
whole.

Since we can’t store the entire stream, one obvious approach is to store a sample

Two different problems:

- Sample a fixed proportion of elements in the stream (say 1 in 10)
- Maintain a random sample of fixed size over a potentially infinite stream

Problem-1:
Example: A search engine receives a stream of queries, and it would like to study the behaviour
of typical users.1 we assume the stream consists of tuples (user, query, time). Suppose that we
want to answer queries such as “What fraction of the typical user’s queries were repeated over
the past month?” Assume also that we wish to store only 1/10th of the stream elements.

Search engine query stream

o Stream of tuples (User, query, time)
o How often did a user run the same query in a single day
o Have space to store 1/10th of query stream

Fixed size sample

Naïve Solution

o Generate a random integer in [0…9] for each query.

o Store the query if the integer is 0, otherwise discard.

Sample users

o Pick 1/10th of users and take all the searches in the sample
o Use a hash function that hashes the user name or user ID uniformly into 10 buckets
Random sapling

Reservoir sampling algorithm

o Store all the first s elements of the stream to S

o Suppose we have seen n-1 elements, and now the nth element arrives ( n > s )
- With probability s/n, keep the nth element else discard it.
- If we picked the nth element then it replaces one of the s elements in the
sample S, picked uniformly at random.

Sliding windows

Useful model

2.5 FILTERING STREAMS

o Identifies the sequence patterns in a stream

o Stream filtering is the process of selection or matching instances of a desired pattern
in a continuous stream of data
o Assume that a data stream consists of tuples
o Filtering steps: (i) Accept the tuples that meet a criterion in the stream, (ii) Pass the
accepted tuples to another process as a stream and (iii) discard remaining tuples

The Bloom Filter Analysis

o A simple space-efficient data structure introduced by Burton Howard Bloom

in 1970.
o The filter matches the membership of an element in a dataset.

Obvious Solution: Hash Table

- But suppose we do not have enough memory to store all of S in a hash table.
E.g., we might be processing millions of filters on the same stream.
A Bloom filter consist of:

 An array of n bits, initially all 0’s

 A collection of hash functions h1, h2, h3, ……., hk.
 Each hash function maps key values to n buckets corresponding to the n bits of the
bit-array.
 A set S of m key values

Blooms filter allows through all stream elements whose keys are in S, while rejecting most of
the stream elements whose keys are not in S.

Illustration of Blooms Filter

Use K-independent hash functions instead of 1.

Fig: Blooms hashing process for the stream (q), K=3, whose hash functions are h1(q), h2(q), and h3(q).

Inserting elements using two hash functions

Searching element with the same two hash functions

Properties of Bloom filter

 No False negative
- If the query was inserted before, bloom filters always return true
 Chance of false positive
- There is a possibility that It can return true for an element which was not inserted
Weaknesses of Bloom filters

 Need full independent hash functions

 Dynamically growing blooms filter is hard
 Best size depends on false positive rate and number of insertions

2.6 COUNTING DISTINCT ELEMENTS IN A STREAM

Finding the number of distinct (Unique) elements in a data stream with repeated elements.

Example:

- Elements might represent IP Addresses of packets passing through router.

- Unique visitors through web sites.
- Elements in a large data set
- Motifs in a DNA dataset
- Elements of a RFID/Sensor network

The Flejolet-Martin (FM) Algorithm

It approximates the number of unique objects in a stream or a Database in one pass

If the stream contains n elements with m of them unique, this algorithm runs in O(n) times and
needs O(log(m)) memory. It gives an approximation for the number of unique objects along
with a standard deviation σ with a maximum error ϵ

Whenever we apply a hash function h to a stream element a, the bit string h(a) will end in some
number of 0’s, possibly none. Call this number the tail length for h(a). Let R be the maximum
tail length of any a seen so far in the stream. Then we shall use estimate 2R for the number of
distinct elements seen in the stream.

FM Algorithm

o Pick a hash function h(a) that maps each of the ‘n’ elements at least log2n bits
o For each stream element a, let r(a) be the number of trailing 0’s in h(a)
o Record R= the maximum r(a) seen.
o Estimate = 2R which is equal to no of distinct elements.
2.7 ESTIMATING MOMENTS

All the Statistical Parameters mean, median, mode are called moments which are used to
estimate or compute the distribution of frequencies of different elements in a stream.

Suppose a stream has elements chosen from a set A of N values

Let mi be the number of times value i occurs in the stream

The kth element is ∑𝑖∈𝐴(𝑚𝑖 )𝑘

Ex: A= {4,5,4,5,3,2,4,5,4,2,4,3}

𝒎𝟒 = 5 𝒎𝟓 = 3 𝒎𝟐 = 2 𝒎𝟑 = 2

Special cases

∑𝒊∈𝑨(𝒎𝒊 )𝒌

0th Moment
The 0th moment is the sum of 1 for each mi that is greater than 0.3 that is, the 0th moment is a
count of the number of distinct elements in the stream

1st Moment
The 1st moment is the sum of the mi’s, which must be the length of the stream. Thus, first
moments are especially easy to compute; just count the length of the stream seen so far

2nd moment
The second moment is the sum of the squares of the mi’s. It is sometimes called the surprise
number, since it measures how uneven the distribution of elements in the stream is.
  To see the distinction, suppose we have a stream of length 100, in which eleven
different elements appear. The most even distribution of these eleven elements would
have one appearing 10 times and the other ten appearing 9 times each. In this case, the
surprise number is 102 + 10 × 92 = 910.
 At the other extreme, one of the eleven elements could appear 90 times and the other
ten appear 1 time each. Then, the surprise number would be 1x 902 + 10 × 12 = 8110.

The Alon-Matias-Szegedy (AMS) Algorithm for Second Moments

Let us assume that a stream has a particular length n
Suppose we do not have enough space to count all the mi’s for all the elements of the stream.
We can still estimate the second moment of the stream using a limited amount of space;
The more space we use, the more accurate the estimate will be.
1. AMS method works for all moments.
2. Gives an unbiased estimate
3. We will just concentrate on the 2nd moment S
4. We pick and keep track of many variables X.
- For each variable X we store X.ele and X.val
 X.ele corresponds to the item i
 X.val corresponds to the count of item i
- Note this requires a count in main memory, so number of X’s is limited.

5. our goal is to estimate

S=∑𝒊(𝒎𝒊 )𝟐

For One Random Variable X

1. How to set X.val and X.ele?

- Assume stream has length n
- Pick some random time t to start
 Let at time t the stream have item i ( we set X.ele=i)
 Then we maintain count c (X.val=c) of the no of i’s in the stream
starting from the chosen time t.
2. Then the estimate of the 2nd moment ∑𝒊(𝒎𝒊 )𝟐 is
S= f(x) = n (2 X c-1)
- Note we will keep track of multiple X’s (X1, X2-----Xn) and our final estimate
will be
Example

Suppose the stream is a, b, c, b, d, a, c, d, a, b, d, c, a, a, b.

The length of the stream is n = 15.
Element Count (mi)

A 5

B 4

C 3

D 3

The second moment for the stream is = ∑𝒊∈𝑨(𝒎𝒊 )𝟐 = 52+42+32+32 = 59

Let X1, X2, and X3 are elements picked randomly (different time stamps) from positions 3rd,
8th, and 13th from the above stream

X1.element = c X1.value= 3
X2.element = d X2.value= 2
X3.element = a X3.value= 2

We can derive an estimate of the second moment from any variable X.

This estimate is = n × (2 × X.value − 1).
= 15 × (2 × 3 − 1) = 75.
= 15×(2 ×2−1) = 45
= 15×(2 ×2−1) = 45

(75+45+45)
Average estimate = = 55
3

2.8 COUNTING 1’S IN A WINDOW

DGIM (Datar, Gionis, Indyk, Motwani) Algorithm

1. DGIM algorithm (Datar, Gionis, Indyk, Motwani) Algorithm.

2. Designed to find number of 1;s in a dataset.
3. This algorithm uses O(log2N) bits to represent a window of N bit.
4. It allows to estimate the number of 1’s in the window with an error of no more
than 50%.
 Components of DGIM Algorithm
1. Timestamp
2. Buckets
- Each bit that arrives has a timestamp for the position at which it arrives
- If the first bit has timestamp 1, the second bit has timestamp 2 and so on.
- The positions are recognized with the window size N ( which are usually taken as
multiples of 2)
- The Windows are divided into buckets consisting of 1’s and 0’s.

Rules for forming the buclets.

There are five rules that must be followed when representing a stream by buckets.
 The right end of a bucket is always a position with a 1.
 No position is in more than one bucket.
 There are one or two buckets of any given size, up to some maximum size.
 All sizes must be a power of 2.
 Buckets cannot decrease in size as we move to the left (back in time).

Example: Fig: Dividing bit-stream into buckets by following the DGIM rules

At the right (most recent) end we see two buckets of size 1. To its left we see one bucket of
size 2. Note that this bucket covers four positions, but only two of them are 1. Proceeding left,
we see two buckets of size 4, and we suggest that a bucket of size 8 exists further left. Notice
that it is OK for some 0’s to lie between buckets. Also, observe from above Fig. that the buckets
do not overlap; there are one or two of each size up to the largest size, and sizes only increase
moving left.
2.9 DECAYING WINDOW

Decaying algorithm allows you to identify most popular elements (trending in other wards) in
an incoming data stream

This algorithm not only tracks most recurring elements in an incoming data stream, but also
discords any random spikes or spam requests that might have boosted an element’s frequency.

In Decaying window algorithm

 We assign a score / weight to every element of the incoming data stream

o For a new element, Reduce the weight of all the existing elements by a constenct
factor k and then assign the new element with specific weight.
 Further, we need to calculate the aggregate sum for each distinct element by adding all
the weights assigned to that element.
o The aggregate sum of decaying exponential weights can be calculated using the
following formula:
𝑡−1

𝑆 = ∑ 𝑎𝑡−𝑖 (1 − 𝑐)𝑖
𝑖=0
Here, t = time stamp
c = small constant
𝑎 = element

 Finally, the element with highest total score is listed as trending or most popular
 Whenever a new element, say at+1 arrives in the data stream, you perform the following steps
to achieve an updated sum

1. Multiply the current sum/score by the value (1-c)

2. Add the weight corresponding to the new element

- In a data stream, consisting of various elements, you maintain a separate sum for each
distinct element.
- For every incoming element, you multiply the sum of the existing elements by a value of
(1-c).
- Further, you add the weight of the incoming element to its corresponding aggregate sum.

Finally, the element with the highest aggregate score is listed as the most popular element.

Example

consider a sequence of twitter tags below

fifa, ipl, fifa, ipl, ipl, ipl, fifa

also let’s say each element in sequence has weight of 1

let c be 0.1

The aggregate sum of each tag in the end of above string will be calculated as below:

Current element is fifa

fifa 1 * (1 – 0.1) = 0.9

ipl 0.9 * (1 – 0.1) + 0 =0.81 (added zero because current tag is other than fifa)

fifa 0.81 * (1 – 0.1) + 1 = 1.729 (added one because current tag is fifa only)

ipl 1.79 * (1 – 0.1) + 0 = 1.5561

ipl 1.5561 * (1 – 0.1) + 0 = 1.4005

ipl 1.4005 * (1 – 0.1) + 0 = 1.2605

fifa 1.2605 * (1 – 0.1) + 1 = 2.315

Current element is ipl

fifa 0 * (1 – 0.1) = 0

ipl 0 * (1 – 0.1) + 1 = 1 (added one because current tag is ipl only)

fifa 1 * (1 – 0.1) + 0 = 0.9 (added zero because current tag is other than ipl)
ipl 0.9 * (1 – 0.1) + 1 = 1.81

ipl 1.81 * (1 – 0.1) + 1 = 2.7919

ipl 2.7919 * (1 – 0.1) + 1 = 3.764

fifa 3.764 * (1 – 0.1) + 0 = 3.7264

In the end of sequence we can see that score of fifa is 2.135 but ipl is 3.7264

So, ipl is more trending than fifa

Real-Time Analytics Platform (RTAP) Applications

Organizations are collecting more data, faster than ever before. And with the number of
connected IoT devices set to reach 24.1 billion by 20301, there’s no sign of slowing. Yet many
companies are struggling to turn these piles of data into insights they can use to grow their
business. This is where real-time analytics can help.

What is RTA: Real-time analytics turns data into insights immediately after it’s collected.
These kinds of insights are used when time is of the essence.

 Operational intelligence, real-time analytics can predict when a device is about to fail,
warning your operations team before it happens.
 Prompt retailers to send mobile promotions to customers when they’re in a store
vicinity.
 Detect credit card fraud before a transaction is completed.

To better understand how real-time analytics works, let’s compare it to traditional analytics, or
batch processing. With the traditional approach, limited sets of historical data are stored and
indexed.

When business users need insights, they query the system. Batch processing is typically used
for routine tasks like generating monthly sales reports or running payroll. While batch
reporting is appropriate for tasks that aren’t time sensitive, others require immediate insights,
such as patient safety monitoring or fraud detection. This is where real-time analytics comes
in.
Types of Real-Time Data Analytics

There are different types of real-time analytics

 On-demand analytics
 Continuous—or streaming—analytics.

On-demand real-time analytics waits for users or systems to request a query and then delivers
the analytic results.

Continuous real-time analytics is more proactive and alerts users or triggers responses as
events happen.

Real-Time Analytics Use Cases

Real-time data allows companies to deliver value-added services and products at the very
moment the customer wants them and to defend against potentially dangerous situations before
they happen.

From retailers and manufacturers to financial services firms and healthcare organizations,
businesses are struggling to keep up with the fast pace of data. Because the value of this data
can evaporate within days, hours, minutes, or even seconds, near-real-time processing is critical
to gaining the most valuable business intelligence.

Real-time analytics addresses many organizational pain points.

 Online retailers are blending transactional and web browsing activity to determine the
next best offer to serve up to a customer.

 Banks are analyzing behaviors to determine fraudulent activity or detect signs that a
customer who works with one of their departments is ready for a pitch from another
department.

 Dynamic pricing

 Risk management

 Call center optimization

 Security are just a few of the processes that can be optimized with real-time analytics.

 Even sports teams make use of streaming analytics to better manage ticketing,
concessions, retail sales, and on-field performance. For example, if a gate gets too
crowded, the organization can immediately send more ticketing and security staff to
that location to help keep wait times down and ensure crowd safety.
Real-Time Sentiment Analysis

It focuses not only on polarity (positive, negative & neutral) but also on emotions (happy,
sad, angry, etc.)helps to understand people’s attitude and behavior

Types of Sentiment Analysis

1. Fine-grained sentiment analysis: This depends on the polarity based. This category can
be designed as very positive, positive, neutral, negative, very negative. The rating is done
on the scale 1 to 5. If the rating is 5 then it is very positive, 2 then negative and 3 then
neutral.
2. Emotion detection: The sentiment happy, sad, anger, upset, jolly, pleasant, and so on come
under emotion detection. It is also known as a lexicon method of sentiment analysis.
3. Aspect based sentiment analysis: It focuses on a particular aspect like for instance, if a
person wants to check the feature of the cell phone then it checks the aspect such as battery,
screen, camera quality then aspect based is used.
4. Multilingual sentiment analysis: Multilingual consists of different languages where the
classification needs to be done as positive, negative, and neutral. This is highly challenging
and comparatively difficult.

Data Extraction: The top trending event and related tweets extracted from particular location
using “Where on Earth Id” (WOEID). WOEID’s are 32-bit identifiers which are non-repetitive
and unique. Once the top trending topic is obtained for a given location, tweets related to these
top trending topics is extracted and stored.

Corpus creation and Pre-processing: It is very important step to clean and pre-process tweets
as it reduces the noise from the data. For doing this, tweets are converted in to a corpus of
words and pre-processing and cleaning of data are done.
The extracted tweets should be freed from
 Punctuation
 white spaces
 special characters such as ‘#’,”@” and numbers
 Stop words such as ‘is’, ‘at’, ‘the’ etc.
 data is converted to lower case for uniformity and better visibility.
 Some of the words such as ‘http’, ‘https’ related to web are also removed.
Stemming and lemmatization is done finally to removes suffixes from words in order to get the
common origin.

This process greatly aid in identifying words with common meaning and that are identical.

Perform Analysis: Real-time sentiment analysis is performed on this pre-processed set of

twitter data by developing and implementing various combinations of machine learning and
lexicon methods
 Document Term Matrix: is a 2-dimensional matrix representation of a data. The terms
(words) are represented in the form of rows and the documents are represented in the
form of columns.
 Labelling using Uni-gram/Bi-gram/N-gram
o By identify the sentiment polarity for each tweet, tweets are classified as
positive, negative or neutral