ITM Web of Conferences 15, 01004 (2017)

DOI: 10.1051/itmconf/20171501004
CMES’17

Modification of Adaptive Huffman Coding for use in encoding

large alphabets
Mikhail Tokovarov1,*
1
Lublin University of Technology, Electrical Engineering and Computer Science Faculty, Institute of Computer Science, Nadbystrzycka 36B,
20-618 Lublin, Poland

Abstract. The paper presents the modification of Adaptive Huffman Coding method – lossless data
compression technique used in data transmission. The modification was related to the process of adding a
new character to the coding tree, namely, the author proposes to introduce two special nodes instead of
single NYT (not yet transmitted) node as in the classic method. One of the nodes is responsible for
indicating the place in the tree a new node is attached to. The other node is used for sending the signal
indicating the appearance of a character which is not presented in the tree. The modified method was
compared with existing methods of coding in terms of overall data compression ratio and performance.
The proposed method may be used for large alphabets i.e. for encoding the whole words instead of
separate characters, when new elements are added to the tree comparatively frequently.

Huffman coding is frequently chosen for implementing

open source projects [3]. The present paper contains the
1 Introduction description of the modification that may help to improve
Efficiency and speed – the two issues that the current the algorithm of adaptive Huffman coding in terms of
world of technology is centred at. Information data savings.
technology (IT) is no exception in this matter. Such an
area of IT as social media has become extremely popular 2 Study of related works
and widely used, so that high transmission speed has
gained a great importance. One way of obtaining high Huffman coding has been developed in 1952 by David
communication performance is developing more Huffman. He developed the method during his Sc. D. study
efficient hardware. The other one is to develop the if MIT and published in the 1952 paper "A Method for
software that would allow to compress the data in such the Construction of Minimum-Redundancy Codes" [1].
a way that would reduce the size of data but not affect its Huffman coding is the most optimal among methods
information content. In other words, to encode the data encoding symbols separately, but it is not always optimal
by the method called lossless data compression. This compared to some other compression methods, such as
term means that the methods of this type allow the e.g. arithmetic and Lempel-Ziv-Welch coding [4].
original data to be perfectly reconstructed from the However, the last two methods, as it has been said, are
encoded message. patent-covered, so developers often tend to use Huffman
Binary or entropy encoding are the most popular coding [3]. Comparative simplicity and high speed due
branches among lossless data compression methods. The to lack of arithmetic calculations are the advantages of
term entropy encoding means that the length of a code is this method as well. Due to these reasons Huffman
approximately proportional to the negative logarithm of coding is often used for developing encoding engines for
the occurrence of the character encoded with the code. many applications in various areas [5].
Simplifying it may be said: the higher probability of the The fact that billions of people are exchanging
character is, the shorter is its code [1]. gigantic amount of data every day stimulated
Several methods of entropy encoding exist, these are development of compression technologies. This topic
the most frequently used methods of this branch: constantly presents significant interest for researches, the
- arithmetic coding, following works may be presented as the examples:
- range coding, Jarosław Duda et al. worked out the method called
- Huffman coding, Asymmetric Numeral Systems, the method was
- asymmetric Numeral Systems. developed on the basis of the two encoding methods:
Arithmetic and Range coding are quite similar with arithmetic and Huffman coding. It combined the
some differences [2], but arithmetic coding is covered by advantages from the two methods:
patent, that is why, due to lack of patent coverage,

*
Corresponding author: [email protected]
© The Authors, published by EDP Sciences. This is an open access article distributed under the terms of the Creative Commons Attribution
License 4.0 (http://creativecommons.org/licenses/by/4.0/).
ITM Web of Conferences 15, 01004 (2017) DOI: 10.1051/itmconf/20171501004
CMES’17

- near accurate symbol probabilities hence better But the compression ratio does not show actual space
compression ratio of arithmetic coding, and capability to saving as besides of encoded message the table with the
fast encoding and decoding of Huffman coding [6]; code-character pairs should be transmitted. For that
- Facebook Zstandard algorithm is based on LZ77 reason, the other index should be introduced. The sent-
dictionary coder and tANS – effective entropy encoding to-original-bits ratio(SOBR) is described in accordance
based on the Huffman method [7]; with the formula below:
- Brotli coding algorithm which is used in most of the
𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛
modern Internet Browsers, such as Chrome, Opera. 𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆 (2)
𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛
Similarly, to the Zstandard it is based on the
combination of LZ77 and modified Huffman coding [8]. The only way to make SOBR equal to DCR is to use
So, it may be clearly seen that despite its long history the one coding tree for all messages, but this tree will not be
Huffman encoding algorithm still presents great interest optimal since the character probabilities may be different
for application. in various messages.

3 Huffman coding explained

3.1 Static Huffman coding algorithm

The concept of Huffman coding is based on the binary
tree. The tree consists of two kinds of nodes:
intermediate nodes, i.e. the nodes having descendant
nodes and the nodes which do not have descendants.
These nodes are called leaves. A character may be stored
only in a leaf node, this condition ensures the character
codes to be prefix-free [1]. It means, that no character
Fig. 1. Huffman tree generated based on the phrase "this is an
has the code, that would be the initial segment of another
example of a Huffman tree".
character's code. As the example of prefix codes, the
following bit sequences may be used: 110101 and 110.
The code 110 is identical to the initial segment of the 3.2 Adaptive Huffman coding algorithm
code 110101, so these two codes may not be decoded
unambiguously. The tree is organized according to the The method of adaptive Huffman Coding (AHC)
following principles [4]: reviewed in the article was proposed by Jeffrey Vitter in
a) any node in the tree may not have a single descendant: his paper published in 1987 [9].
either two or none; The algorithm working during creating the tree may
b) each node in the tree has the number assigned to it. be observed in Figure 2.
This number is called weight. Depending on the type of
the node its weight may have the following meanings:
- if the node is a leaf, the weight value is equal to the
number of times the character stored in the leaf occurs in
the message sent;
- if the node is an intermediate node its weight value is
equal to the sum of its descendants' weights.
c) the weight of the right descendant should be not less
than the weight of the left descendant. Fig. 2. The algorithm of AHC, example for encoding the word
The codes for every character are defined as the path "abb".
in the binary tree from the root to the leaf containing the
character (See Figure 1), e.g. the blank space character This method is based on the same principles as the
which is the most frequent character in the tree has the static method plus the following extensions [9]:
code 111, and the 'p' character, which occurs only once a) every node in the tree has its key number, the key
has the code 10011 [4]. numbers are arranged in the following way:
The Figure 1 presents the tree constructed in - maximum value of the key number in the tree may be
accordance with static method of Huffman coding. The calculated as:
main feature of it is that the tree is constructed before the
transmission is started on the basis of analysis of the 𝑘𝑘𝑘𝑘𝑘𝑘𝑘𝑘𝑘𝑘𝑘𝑘 𝑘 𝑘 𝑘 𝑘𝑘𝑘𝑘𝑘𝑘𝑘𝑘𝑘𝑘𝑘𝑘𝑘𝑘𝑘𝑘𝑘𝑘𝑘𝑘𝑘𝑘𝑘𝑘𝑘𝑘𝑘𝑘𝑘𝑘𝑘𝑘𝑘𝑘𝑘𝑘𝑘𝑘𝑘𝑘𝑘𝑘 𝑘 𝑘 (3)
probabilities of separate characters in the whole message
- the root has the largest key number;
[1].
- an ancestor has larger number than any of its
The data compression ratio(DCR) is described by the
descendants;
formula (1):
- the right descendant should have larger key number
𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛 than the left descendant;
𝐷𝐷𝐷𝐷𝐷𝐷 𝐷 (1)
𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛 b) after input of any character the tree is updated;

2
ITM Web of Conferences 15, 01004 (2017) DOI: 10.1051/itmconf/20171501004
CMES’17

c) a special leaf node called NYT (not yet transmitted) is

used for both indicating the place for a new character Listing 1. Pseudocode presenting updating tree of AHC.
and for signalizing that the new character is obtained, its UpdateTree(character ch)
key has the least value in the tree, its weight always 1 if ch is not found in the tree
equals to zero; 2 make a newLeafNode as the right descendant of the
d) a set of nodes with equal weight values is called a block. oldNYTNode;
The encoding algorithm is described in Listing 1. The 3 newLeafNode's keyValue := oldNYTNode's
algorithm contains only the process of updating the tree keyValue-1;
and does not consider communication. If the 4 make a newNYTNode as the left descendant
communication algorithm based on the AHC is used, the of the oldNYTNode
process becomes more complicated. The idea may be 5 newNYTNode's keyValue := oldNYTNode's
presented as follows: keyValue-2;
- transmitter and receiver have identical trees which are 6 NYTNode := newNYTNode;
updated in accordance with the algorithm described in 7 end if
the Listing 1; 8 activeNode := node containing ch;
- the communication operates in the way presented in the 9 do
Listing 2. 10 find the nodeWithTheLargestKeyNumber
As it may be seen from the Listing 2, in case of AHC in the activeNode's block;
not only the codes of the characters are transmitted. 11 if the nodeWithTheLargestKeyNumber's keyNumber
Auxiliary codes such as the code of NYTNode and ASCII > activeNode's keyNumber
12 swap nodeWithTheLargestKeyNumber
codes of the new-coming characters are being
and activeNode;
transmitted as well. The necessity to transmit auxiliary
13 swap nodeWithTheLargestKeyNumber's
codes negatively influences SOBR causing it to increase.
keyNumber and activeNode's keyNumber;
Due to this fact overall data saving decreases. But along //key numbers should not change the place
with higher degree of complexity compared to static 14 end if
Huffman coding ACH may still present interest as 15 activeNode's weight := activeNode's weight+1;
lossless data compression technique after implementing 16 activeNode := activeNode's ancestor;
the modifications described in the next chapters. 17 while activeNode <> root

Listing 2. Communication with the use of data compression

3.3 Encoding words instead of separate characters based on AHC.
The title of the article contains the phrase "large Transmitter Receiver
alphabet". But the chapter is focused at encoding entire Transmit (character ch) Receiving is performed in the
words. What is the connection between these two facts? 1 if ch is found in the tree stream, i.e. in infinite loop.
The idea is that in the method proposed, the words are 2 send the code of ch Receive()
treated as separate characters, so the leaves of the coding bit by bit; 1 while (true)
tree store not characters, but complete words, so these 3 else 2 receive one bit from
words are treated as separate characters in a large 4 send the code of the transmitter,
alphabet. The author does not claim, that this idea NYTNode bit by bit; 3 add received bit to
belongs to him. This technique is quite well known and 5 send the ASCII code bitSequence;
was presented in many works [5, 10]. of ch; 4 if bitSequence
The greatest success may be achieved in the case of 6 end if leads to a leaf node
applying this method for encoding the words of an 7 UpdateTree(ch); 5 obtain the character
analytic language. An analytic language is the type of ch stored in the node;
language where grammatical relationships are 6 add ch to decoded
established by using strict word order, prepositions, message;
postpositions, particles and special auxiliary words that 7 UpdateTree(ch);
do not have individual meaning and only indicate some 8 set bitSequence
empty;
grammatical categories. Analytic languages do not have
9 end if
extensive systems of conjugation and declension as
10 if bitSequence leads
synthetic languages do [11].
to NYTNode
In many cases a word in a synthetic language may 11 receive 8 bit;
have several forms which are treated as individual words 12 convert received 8
by encoding algorithm. This approach would cause the bit to character ch;
coding tree to be excessively extensive. It is worth to 13 add ch to decoded
note that the statistics show that around 95% of all message;
common English texts may be covered by 7000 words 14 UpdateTree(ch);
[11, 13]. The situation becomes even more optimistic 15 set bitSequence
when communication in social networks and mass media empty;
is considered. 16 end if
17 end while

3
ITM Web of Conferences 15, 01004 (2017) DOI: 10.1051/itmconf/20171501004
CMES’17

To prove the feasibility of the method the term of coming word is attached to and for sending the signal
information entropy should be mentioned. This term has meaning that the ASCII code of the new-coming signal
several definitions: is going to be sent. This fact provides the opportunity for
- measure of unpredictability of the state; optimising the algorithm.
- expected (mean) value of information contained in
a message.
The value of entropy is calculated by the following 4 Modification
formula:
Entropy of i-th character in a message: 4.1 Introduction of NCW node
𝐼𝐼𝑖𝑖 = −𝑙𝑙𝑙𝑙𝑙𝑙𝑚𝑚 𝑝𝑝𝑖𝑖 (4) The NYT node should only act as the pointer for the
new-coming word. Its weight still should be 0 and its
Average entropy of a message: key number should have the least value in the tree. The
𝐻𝐻 𝐻 ∑𝑛𝑛𝑖𝑖𝑖𝑖 −𝑝𝑝𝑖𝑖 𝑙𝑙𝑙𝑙𝑙𝑙𝑚𝑚 𝑝𝑝𝑖𝑖 = ∑𝑛𝑛𝑖𝑖𝑖𝑖 −𝑝𝑝𝑖𝑖 𝑙𝑙𝑙𝑙𝑙𝑙𝑚𝑚 𝐼𝐼𝑖𝑖 (5) new node NCW should be introduced to the tree. Its
initial weight should be equal 0. This node should be
where: pi - probability of i-th character in the message; used for sending the "new-coming word" signal. The
n - number of unique characters in the message; introduced NCW node should be treated as a usual leaf
m - logarithm base, usually taken equal to 2, as binary node, i.e. after sending the bit sequence corresponding to
system is used in computer technics. the NCW node, the procedure described in the lines 8-17
Practically it may be stayed, that the average entropy of the Listing 1 should be carried out for the NCW node.
of a message is equal to the least possible average code The Listing 3 presents the modified algorithm.
length of the characters contained in the message [1].
Listing 3. Modified algorithm.
It is well known, that amongst discrete distribution
with equal number of states the uniform distribution has Transmitter Receiver
the maximum value of entropy [12]. Every state of the Transmit (character ch) Receiving is performed in the
uniform distribution has the same probability equal to 1 if ch is found in the tree stream, i.e. in infinite loop.
1/n, where n is the number of states, which yields the 2 send the code of ch bit Receive()
entropy: by bit; 1 while (true)
3 else 2 receive one bit from the
𝐻𝐻𝑢𝑢𝑢𝑢𝑢𝑢𝑢𝑢𝑢𝑢𝑢𝑢𝑢𝑢 = 𝑙𝑙𝑙𝑙𝑙𝑙2 𝑛𝑛 (6) 4 send the code of transmitter,
NCWNode bit by bit; 3 add received bit to
To estimate the maximum entropy, the maximum 5 bitSequence;
number of words encoded should be defined. It has been updateTree(NCWNode); 4 if bitSequence leads to a
decided to take the max number of words equal to 6 send the ASCII code of leaf node
16384, which yields the maximal entropy equals to 14 ch; 5 obtain the character
bits. The max word number is taken because the 7 end if ch stored in the node;
practical number of stored words may be higher than 8 UpdateTree(ch); 6 add ch to decoded
7000 because of some capitalized words, abbreviations, message;
mistakes and user-defined words. To roughly estimate 7 UpdateTree(ch);
the compression ratio, the average length of English 8 set bitSequenc
word is used, its value approximately equals to 4 letters. empty;
In case if ASCII coding is taken into consideration the 9 end if
average length in bits equals to 32. Based on this value 10 if bitSequence leads to
the average compression ratio equals to 14/32 ≈ 0.43. NCWNode
This estimation is fairly promising as average 11updateTree(NCWNode);
compression ratio for separate-character AHC is 12 receive 8 bit;
about 0.55 [4]. 13 convert received 8 bit
As it is stated in the previous chapter, sent-to-original to character ch;
14 add ch to decoded
bit ratio tends to be significantly higher than data
message;
compression ratio for complete-word AHC. This effect is
15 UpdateTree(ch);
especially noticeable during the phase of initial building
16 set bitSequence empty;
the coding tree, when new words are coming especially 17 end if
frequently. There are two factors that affect SOBR in 18 end while
this case: sending the complete ASCII codes of the new-
coming words and sending the NYT bit sequence. The Figure 3 presents the difference between
Precoded dictionaries may be used to decrease the modified and non-modified method. The phrase
influence of first factor. This possibility is not "A friend in need is a friend indeed" was used for the
considered in the current paper. test. As it may be noticed the weight of the NCW node
However, the second factor, i.e. sending the NYT bit is equal to the number of the unique leaf nodes.
sequence will be optimized within the frame of this
research. As it is described in the chapter 2 the NYT is
used for both indicating the place in the tree the new-

4
ITM Web of Conferences 15, 01004 (2017) DOI: 10.1051/itmconf/20171501004
CMES’17

Fig. 4. Example of forgetting function operation.

Fig. 3. Comparison of full-word AHC trees. Left – two-
purpose NYT node, right – separate NCW and NYT nodes.
5 Experiment and results
4.2 Forgetting
The tests were conducted on the data set containing more
It is a quite frequent fact in real social media application, than million characters. The data set was composed on
when a user makes some mistakes in the text. In the case the basis text collected from various sources: news posts,
of the complete-word Huffman algorithm it would cause social network comments and private correspondence.
the coding tree to be overgrown and not optimal, since Use of forgetting function was completely feasible as
these mistyped words are used very rarely, but keep the many rare words, such as proper names and mistypes
place in the tree, causing the entropy to be larger. The appeared. The results have proven that another
same thing may be stated about rare words. Another application of complete-word AHC may be composing
problem that may raise is overflowing of dynamically a dictionary for dictionary-type coding. Table 1 presents
allocated memory. In the current application some frequent words and symbols found in the test texts.
dynamically, allocated arrays are used instead of linked
Table 1. The most frequent words and symbols used in the
lists due to their better performance rates. To deal with
text.
these problem, the algorithm of forgetting should be
introduced. Word Weight Code
This algorithm is based on the estimation of the
he 937 101001001
relevance function of the stored words. The relevance
function may be defined as Euclidean norm: have 887 100111001
as 866 100110101
𝑊𝑊𝑊𝑊 = √𝑤𝑤𝑤𝑤𝑤𝑤𝑤𝑤𝑤𝑤𝑤 2 + 𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎 2 (7)
be 866 100110100
where: agingFactor is the value characterizing how long are 815 100011101
cycles ago the word was used last time; weight is the from 796 100011001
number of word's appearances in the text, the more the
his 793 100011000
weight, the more often appears the word in the text.
The forgetting function is based on the algorithm has 781 100010101
presented in Listing 4. at 770 100010011
Listing 4. Forgetting function. Trump 755 100000000
not 752 1111111110
1 if numberOfStoredWords > thresholdWordNum
2 Sort the array of the leaf nodes basing on WF in The next issue that arose during the tests and
descending order; comparison of the algorithms was the selection of
3 while numberOfStoredWords > desiredWordNum features for analysis. The overall sent-to-original bits
4 delete the last word from the leaf nodes array;
ratio has been chosen first. The Figure 5 presents the
5 end while
comparison of SOBR for the three implemented
6 end if
methods: separate-character AHC, unmodified complete-
It is reasonable to set thresholdWordNum close to the word AHC and modified complete-word AHC.
max number of stored words and desiredWordNum – It may be clearly seen from the picture that even
approximately 10-15% less to ensure periodical unmodified complete-word AHC demonstrates better
"cleaning" of the tree. sent-to-original bit ratio, as it was stated in the chapter
The Figure 4 presents the example of forgetting 3.3. The modified method allows to decrease the SOBR
function operation. The threshold number of words is even more. The end difference between overall SOBR
equal to 16000 and the desired number of words is values for modified and unmodified methods equals to
15000. 2.3 %.

5
ITM Web of Conferences 15, 01004 (2017) DOI: 10.1051/itmconf/20171501004
CMES’17

- better SOBR comes at a cost: the algorithm needs more

memory to store the tree, the search of the node in the
tree is slower due to its size, initial sent-to-original bits
ratio is higher than this of separate-character AHC.
The implemented modification allowed to further
improve the SOBR of complete-word AHC. The tests
have shown that:
- sent to original bits ratio is 15.1 percentage points less
Fig. 5. Comparison of SOBR for the implemented AHC methods. compared to separate-character AHC method;
- the proposed modification proved to be more effective
Another feature that may help to compare the two during the periods when large number of words is being
complete-words methods is specified sent-to-original added to the tree, and less when the number of new
bits ratio (SSOBR). Its value may be computed as first words is decreasing. Possible solution, that would be
order divided difference of SOBR: useful in that case, would be using the modified
algorithm while the tree is being formed and then
𝑁𝑁𝑁𝑁𝑁𝑁𝑁𝑁𝑖𝑖 −𝑁𝑁𝑁𝑁𝑁𝑁𝑁𝑁𝑖𝑖𝑖𝑖
𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑖𝑖 = (8) deleting the NCW node by the means of the function
𝑁𝑁𝑁𝑁𝑁𝑁𝑁𝑁𝑖𝑖 −𝑁𝑁𝑁𝑁𝑁𝑁𝑁𝑁𝑖𝑖𝑖𝑖
used for "forgetting" and using unmodified algorithm.
where: NBSM is the number of bits in the sent message;
NBOM is the number of bits in the original message;
i is the number of step.
References
The value of the denominator corresponds to the step 1. D.A. Huffman, Proceedings of I.R.E, pp. 1098–
or the number of bits the specified SOBR is computed 1101, (1952)
per. In the current research the step was chosen to be
equal to 1 Kbyte. The Figure 6 presents the differences 2. G. Nigel N. Martin, Video & Data Recording
of SSOBR of unmodified and modified AHC. Conference, pp. 612-620, (1979)
3. B. Ryabkot, Data Compression Coding Conference
Proceedings, pp 246-252, (2004)
4. J. Van Leeuwen, ICALP, pp. 382-410, (1976)
5. J. Lee, MIT Undergraduate Journal of Mathematics,
pp. 122-130, (2007)
6. J. Duda; K. Tahboub; N.J. Gadgil; E.J. Delp, Picture
Coding Symposium (PCS), pp. 65-69, (2015)
7. https://code.facebook.com/posts/1658392934479273
/smaller-and-faster-data-compression-with-
Fig. 6. Differences between SSOBR of unmodified and modified zstandard/ - last access 07.07.2017
AHC methods. 8. J. Alakuijala, Z. Szabadka, Internet Engineering
Task Force (IETF), (2016)
The Figure 6 shows that the difference between
SSOBR of the both methods tends to be more in the 9. J. S. Vitter, Journal of the ACM, 34(4), pp 825–845,
periods when a lot of new words is added to the tree (up (1987)
to 6 percentage point). However, it might have quite low 10. M.B. Baer, 2013 IEEE International Symposium on
and even negative values, as in case of absence of new Information Theory Proceedings (ISIT), pp. 1749-
words unmodified AHC method becomes more optimal. 1753, (2013)
11. V.V. Bochkarev, A.V. Shevlyakova, V.D. Solovyev,
Social Evolution & History, Vol.14, number 2, pp.
6 Conclusion
153-175 (2015)
The comparison of single-character AHC and complete 12. H. Yokoo, 2016 International Symposium on
words AHC have been conducted, the research has Information Theory and Its Applications (ISITA),
proven the following: (2016)
- overall sent-to-original bits ratio of complete word 13. http://www.ravi.io/language-word-lengths - last
AHC is approximately 12.8 percentage points lower than access 07.07.2017
separate-words AHC one;