A PERSONALIZED RECOMMENDER SYSTEM FOR WRITING

IN THE INTERNET AGE

M. C. Puerta Melguizo

, O. Muñoz Ramos

, T. Bogers

, L. Boves

and A. van den Bosch

Department of Language and Speech, Radboud University, P.O. Box 9103, 6500 HD Nijmegen, The Netherlands

ILK/Language and Information Science, Tilburg University, P.O. Box 90153 NL 5000, LE Tilburg, The Netherlands

Keywords: Proactive recommender systems, writing stages, information seeking, long-term memory.

Abstract: With the advent of Internet, writing and finding information to plan and structure the text have become

increasingly intertwined. We think that it is necessary to develop systems able to support the task of finding

relevant information, without interfering with the writing process. The Proactive Recommender System À

Propos is being developed in order to support writers in finding relevant information during writing. We

present our research findings and raise the question whether the tendency to interleave (re)search and

writing implies a need for developing more comprehensive models of the cognitive processes involved in

writing scientific and policy papers.

1 INTRODUCTION

Writing professional documents (e.g. scientific

papers, user manuals, etc) is complex. The most

widely influential model of writing is the one

proposed by Hayes and Flower (1980). Although in

this model the processes considered are the ones

involved in writing with pen and paper, virtually all

software systems that have been design seem to

build on the concepts developed in this model. We

will start by introducing the model of Hayes and

Flower and how the use of computers and the

internet has changed the way we write. We will

finish by presenting research we are performing in

order to develop a Proactive Recommender System:

À Propos. Our research is based on the conviction

that in order to design better tools for writing it is

important to understand the cognitive processes

involved in writing and searching information.

2 THE COGNITIVE PROCESSES

OF WRITING

According to Hayes and Flower (1980), writing

happens in three stages: Planning, Translating, and

Reviewing. During Planning ideas are generated and

arranged into a coherent structure. Planning involves

retrieving domain knowledge from the writer’s

Long-Term memory (LTM). During translating

writer’s plans are transformed into sentences. In the

Reviewing stage the writer evaluates the relation

between the text written so far and the linguistic,

semantic and pragmatic aspects that would best

serve the writing goal. Reviewing involves reading

and editing errors or weaknesses in the text. The task

environment includes everything outside the writers'

mind that can influence the writing task including

the text produced so far and the so called rhetorical

problem (the writing assignment, the specification of

topic and the audience). In the writer’s LTM are

stored the writer’s knowledge about the topic, the

knowledge of sources based on literature search, the

writing plans and the knowledge about the audience

who will read the work.

Hayes (1996) extended the model and

emphasized the role of working memory, as well as

socio-cultural and motivational aspects in writing.

Furthermore, the task environment is divided into

social and physical contexts. According to Hayes the

social environment needs to be considered because

writing is a social activity, and consequently, the

way a text is written is affected by several cultural

conventions and the audience it is meant for. In the

physical environment, the composing medium or tool

used to write has been added to the text produced so

far. Actually, variations in the medium seem to lead

to differences in the way people carry out the writing

task. For example, Haas (1996) found that writers

335

C. Puerta Melguizo M., Muñoz Ramos O., Bogers T., Boves L. and van den Bosch A. (2008).

A PERSONALIZED RECOMMENDER SYSTEM FOR WRITING IN THE INTERNET AGE.

In Proceedings of the Tenth International Conference on Enterprise Information Systems - HCI, pages 335-338

DOI: 10.5220/0001682103350338

 SciTePress

tend to plan more and review at a more general level

when they write on paper than when using a word

processor. These results suggest that the introduction

of computer tools seems to force users to change the

processes they use. However, still a lot of research

needs to be done in order to explore how the

composing medium affects the writing process.

3 WRITING IN THE INTERNET

AGE

Current models of writing assume that knowledge

about the topic of the text is mainly stored in the

writer’s neural LTM. The reality of writing

professional texts however shows that writers almost

invariably need to look for additional external

information while writing. And with the advent of

the Internet more frequently than ever, writing is

now interleaved with searching for information. Yet,

seeking for information is difficult and time

consuming. Keyword-based search is still inefficient

and relevant information may be missed. Also

considerable time is spent interacting with low-

precision search engines. Consequently, the time in

which the author is away from creating the

document can have a negative impact on the total

time spent and on the quality of the text.

Furthermore, we question whether continuously

switching between writing and searching is efficient,

and whether it tends to result in the best possible

quality of the texts. Finally, we think it is necessary

to design tools that support writing and help users to

retrieve relevant information.

4 À PROPOS

A Proactive Recommendation System (PRS) relieve

authors from explicit search and switching between

applications by means of searching information

accurately and recommending this information in a

proactive manner. For example, Watson (Budzik and

Hammond, 1999) performs automatic Web searches

based on text being written or read. A problem with

current PRSs is that they are developed as search

tools and do not take into account the specific

characteristics of the writing task.

Our goal is to develop a PRS for writers in a

professional environment: À Propos. The

architecture is based on a client-server architecture.

The client runs on the user's computer and monitors

user’s activity constantly. À Propos proactively

submits queries based on the user and group profiles

in combination with what the user is currently typing

or reading. The server consults the relevant

information sources, and returns the search results to

the client. A more detailed description of the

system’s architecture can be found in (Puerta

Melguizo et al., 2007a) where the role of the

different components of the system such as

observers, filters and gatekeepers is explained. In the

User Interface the results of the search are presented

in a semi-transparent window located in the bottom

right of the screen (see Figure 1). The window

contains URLs related to what the user is typing. As

the user moves the cursor over the references, the

URLs become fully visible and active. On clicking

the required URL, the user accesses the

corresponding paper from the digital library. The

information in the window changes depending upon

the text that is being input and new queries that are

created. To develop À Propos two main issues are

being researched. First, in order to present highly

relevant information, appropriate filtering techniques

need to be developed. Second, procedures to identify

the different writing stages and related information

needs must be created in order to design an

appropriate user’s interface. The researches

performed for both issues are discussed below.

Figure 1: The user’s interface.

4.1 Selecting and Presenting Relevant

Information

The acceptance of any PRS hinges on the relevance

and accuracy of the suggested information. Quality

recommendations should be both on topic and

personalized. To increase the topicality of the

recommendations one can use detailed personalized

taxonomies integrated in an easily expandable, yet

robust IR model to retrieve the initial list of

documents. We are investigating personalization on

user and group level.

ICEIS 2008 - International Conference on Enterprise Information Systems

336

4.1.1 User Personalization

When personalizing results, we consider the user’s

interests and expertise. From these data we build a

profile of terms important for the user which is used

to re-rank the initial recommendations and

suggestions with more matching profile terms get

promoted to the top of the list. Three different

sources of information are considered for inclusion

in the user profile: past selections from the list of

recommendations, user’s past documents, and the

PRS also allows users to enter informational queries

manually.

4.1.2 Group Personalization

À Propos aims to perform group personalization by

identifying the expertise in different topics of the

members of a group. The user’s own documents and

profile are seeing as an expertise fingerprint of that

user. We can then use taxonomies (e.g. the ACM

hierarchy) to represent the hierarchy of topics for

which we want to quantify a group members’

expertise. By collecting an adequate number of

documents for each topic we can extract the

representative terms and construct topic fingerprints.

The next step is to match these topic fingerprints

with the user’s expertise profile by calculating the

term overlap. This way we can calculate the

expertise of each group member on the different

topic areas and also find out which group members

are experts in the topic of the user’s active

document. Knowledge of the distribution of

expertise over the group can then be used for

personalization. For instance, the recommendation

of a document by an expert on the topic should be

considered as more reliable and have a significant

influence on the final re-ranking (Bogers and Van

den Bosch, 2006). Group personalization could also

be used to recommend documents that were not even

in the initial recommendation list. Expertise

fingerprints can also be compared to each other and

used to suggest related topics to the user to provide

for a more serendipitous experience. Our experience

suggests that serendipity is especially important in

the earliest phases of planning.

4.2 The Problem of Interrupting the

Stages of Writing

One problem with presenting proactive information

is that it can interrupt the ongoing writing task. The

interruption can also be more disturbing and

distracting in specific stages of the writing process.

Consequently, the effects of interruptions during

different writing stages need to be considered.

Deshpande et al., (2006) found that writers need to

look for extra information especially during

planning and reviewing. Consequently, we decided

to study the effects of presenting proactive

information during these stages (Puerta Melguizo et

al., 2007b)

4.2.1 Presenting Proactive Information

during Planning Tasks

To simulate the stage of Planning, participants were

told that to write essays, they had to start by writing

an outline of the major points and order in which

they would be introduced in the essay. The writing

outline was the planning task. Participants wrote the

planning outlines: 1) without PRS and no option of

looking for extra information, 2) without PRS and

the option of getting information by actively

searching information in the Web, 3) with

presentation of proactive relevant information by our

PRS, and 4) with presentation of non-relevant

information by our PRS

The PRS did not seriously impair time

performance. Furthermore, when relevant

information was presented proactively, the quality of

the writing plan was significantly better and

participants introduced more information than in the

other conditions. The results of this experiment also

show that active search initiated by the user resulted

in a lower quality of the information found and a

worse written text.

4.2.2 Presenting Proactive Information

during Reviewing Tasks

Participants performed two editing tasks: spelling

corrections and filling in factual information in the

text. Participant performed both reviewing tasks

under three conditions: 1) without PRS and the

option of getting information by actively searching

information in the Web, 2) with presentation of

proactive relevant information by our PRS 3) with

presentation of non-relevant information by our

PRS.

Again, the presentation of proactive information

did not impair time performance. Furthermore, the

time spent in looking for new relevant information

was shorter when the PRS presented relevant

information than when participants searched for the

information actively. The information seeking time

was even longer when non-relevant information was

presented proactively. In this case, after assessing

that the information by the PRS could not help in

A PERSONALIZED RECOMMENDER SYSTEM FOR WRITING IN THE INTERNET AGE

337

completing the editing task, participants started an

active search. This result emphasizes the importance

of developing appropriate search profiles and filters

as described above. Finally, the quality of the editing

tasks was also significantly better when proactive

relevant information was presented showing once

more, that active search initiated by a user is less

effective.

4.3 An External Long-Term Memory

Virtually all writing research has been conducted in

settings in which the LTM from which participants

could ‘get information’ was limited to their own

brain. However, the advent of Internet is already

affecting the way people consider and use LTM and

now is becoming more important to know how to

find information than to memorize information in

the first place. However, accessing information in

the Internet is not without problems. Knowing less,

while searching more makes more difficult to assess

the importance of found information and to integrate

it in a coherent framework. A PRS could be able to

support the decisions about the relevance of the

results returned from a query and be used as an

addition to the writer’s neural LTM. Furthermore,

we think it is neccesary to develop a new model of

cognitive writing processes in which the external

LTM that the WWW and other databases conforms,

needs to be included as an important part of the

physical environment.

5 CONCLUSIONS

In this paper we presented the PRS À propos. This

system is in development and aims at supporting

writers in the difficult task of finding appropriate

relevant information during writing.

First, we present the efforts we are performing in

order to develop adequate group and personal

profiles that make sure the information presented by

the system is relevant to the writer and to the

specific piece of text is being written. We also

describe the studies we performed in order to

explore the effects of presenting proactive relevant

information when writers are planning and

reviewing text. From our experiments we could

conclude that the user’s interface of the PRS does

not negatively interrupt the task of writing. And

even more important, when relevant information is

presented, the quality of the writing text

significantly improves in comparison with the

situations in which the user actively seeks for

information. Furthermore, the results of our

experiments with proactive presentation of

information suggest that professionals are willing to

accept unsolicited pop-up windows and similar

interrupts if the information that they are alerted to

by those interrupts is relevant for the completion of

their (writing) task.

REFERENCES

Bogers, T. and Van den Bosch, A., 2006. Authoritative

Re-ranking of Search Results. In Proceedings of the

28th European Conference on Information Retrieval

(ECIR 2006), vol. 3936 of Lecture Notes on Computer

Science, pp. 519-522. Springer Verlag, April 2006.

Budzik, J. and Hammond, K., 1999. Watson: Anticipating

and Contextualizing Information Needs. In

Proceedings of the 62nd Annual Meeting American.

Society for Information Science, 727-740.

Deshpande, A., Boves, L. and Puerta Melguizo, M.C.,

2006. À propos: Pro-active personalization for

professional document writing. In SigWriting, 10

International Conference of the EARLI Special

Interest Group on writing. Antwerp, Belgium.

Haas, C., 1996. Writing Technology Studies on the

Materiality of literacy. Lawrence Erlbaum Associates,

Hillsdale New Jersey.

Hayes, J.R., 1996. A new framework for understanding

cognition and affect in writing. In C.M. Levy and S.E.

Ransdell (Eds.). The science of writing: Theories,

methods, individual differences, and applications (pp.

76-97). Lawrence Erlbaum Associates, Hillsdale New

Jersey.

Hayes, J.R. and Flower, L.S., 1980. Identifying the

organization of writing processes. In L.W. Gregg and

E.R. Steinberg (Eds.). Cognitive processes in writing

(p. 3-30). Lawrence Erlbaum Associates, Hillsdale

New Jersey.

Puerta Melguizo, M. C., Bogers, T., Deshpande, A.,

Boves, L., and Van den Bosch, A., 2007a. What a

proactive recommendation system needs - relevance,

non-intrusiveness, and a new long-term memory. In

Proceedings of the ICEIS: 9th international

conference on enterprise information systems, 86-98.

Puerta Melguizo, M.C., Boves, L, Deshpande, A., and

Muñoz Ramos, O., 2007b. A Proactive

Recommendation System for Writing: Helping

without Disrupting. In W-P. Brinkman, D-H. Ham and

W. Wong (Eds.). ECCE 2007: European Conference

on Cognitive Ergonomics, 89-95.

ICEIS 2008 - International Conference on Enterprise Information Systems

338